CN102880688B - A kind of method for webpage is estimated, device and equipment - Google Patents
A kind of method for webpage is estimated, device and equipment Download PDFInfo
- Publication number
- CN102880688B CN102880688B CN201210343159.4A CN201210343159A CN102880688B CN 102880688 B CN102880688 B CN 102880688B CN 201210343159 A CN201210343159 A CN 201210343159A CN 102880688 B CN102880688 B CN 102880688B
- Authority
- CN
- China
- Prior art keywords
- webpage
- group
- model
- information
- target sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Information Transfer Between Computers (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
It is an object of the invention to provide and a kind of determine the method for webpage assessment models, device and equipment based on transfer learning method.The method according to the invention, including: obtain one group of target sample group of webpage assessment models to be set up and corresponding least one set sample for reference group;According to described target sample group and described least one set sample for reference group, it is thus achieved that in described least one set sample for reference group part or all refer to the sample group weight adjustment information relative to described target sample group;Described weight adjustment information and described predetermined training method corresponding to described target sample group, each sample for reference group perform migration operation, it is thus achieved that the webpage assessment models of described target sample group.It is an advantage of the current invention that: can be rapidly achieved convergence, it is established that stable assessment models, therefore, it is possible to be well applicable to the process of mass data, the assessment for the webpage of current magnanimity can obtain good effect.
Description
Technical field
The present invention relates to field of computer technology, particularly relate to a kind of method for webpage is estimated, device and equipment.
Background technology
The life of people is day by day incorporated along with the Internet, advertisement on the Internet has become as enterprise and carries out the important means of product/service marketing, and corresponding, how user in the urgent need to understanding the effect of advertisement putting, to adjust advertisement serving policy accordingly according to throwing in effect.But, obtained all data are set up model every time, not only inefficient, and also the deviation of its model and variation are all bigger.Therefore, it is established that assess the assessment models of advertising message that will throw according to given data, that is adopt moving method to set up the urgent needs that assessment models is many people.
But, due to the rapid expanding of internet scale, Various types of data amount is sharply incremented by, and advertising message is no exception, is namely mass data easily.And existing migration algorithm requires over to be iterated operating to each data and realizes, when mass data, even if being operated by the learning training of long period, it is also difficult to reach a stable convergence state.
So, utilize data with existing, foundation how can be optimum can be rapidly achieved the model of convergence in mass data situation, are current problem demanding prompt solutions.
Summary of the invention
It is an object of the invention to provide and a kind of determine the method for webpage assessment models, device and equipment based on transfer learning method.
According to an aspect of the present invention, it is provided that a kind of method for webpage is estimated, wherein, described webpage includes multinomial feature relevant information, said method comprising the steps of:
A obtains one group of target sample group of webpage assessment models to be set up and corresponding least one set sample for reference group, and wherein, the least one set sample for reference group of described target sample group and correspondence thereof includes multiple webpage respectively;
B is according to described target sample group and described least one set sample for reference group, it is thus achieved that in described least one set sample for reference group part or all refer to the sample group weight adjustment information relative to described target sample group;
The c described weight adjustment information corresponding to described target sample group, each sample for reference group and described predetermined training method perform migration operation, it is thus achieved that the webpage assessment models of described target sample group;
One or more webpages to be assessed are estimated by d based on described webpage assessment models.
According to another aspect of the present invention, additionally providing a kind of webpage apparatus for evaluating for webpage is estimated, wherein, described webpage includes multinomial feature relevant information, and described webpage apparatus for evaluating includes:
First acquisition device, for obtaining one group of target sample group of webpage assessment models to be set up and corresponding least one set sample for reference group, wherein, the least one set sample for reference group of described target sample group and correspondence thereof includes multiple webpage respectively;
Weight Acquisition device, for according to described target sample group and described least one set sample for reference group, it is thus achieved that part or all refer to the sample group weight adjustment information relative to described target sample group in described least one set sample for reference group;
Model acquisition device, performs migration operation for the described weight adjustment information corresponding to described target sample group, each sample for reference group and described predetermined training method, it is thus achieved that the webpage assessment models of described target sample group;
Apparatus for evaluating, for being estimated one or more webpages to be assessed based on described webpage assessment models.
According to another aspect of the present invention, additionally providing a kind of computer equipment, wherein, this computer equipment includes described webpage apparatus for evaluating.
Compared with prior art, the invention have the advantages that
1) need each data are iterated operation according to the migration operation of the present embodiment, so convergence can be rapidly achieved, setting up stable assessment models, therefore, it is possible to be well applicable to the process of mass data, the assessment for the webpage of current magnanimity can obtain good effect.
2) by performing Data Migration or model migration operation, based on current existing data or model, corresponding with target sample group, rational assessment models can be set up, so that each webpage belonging to target sample group to be estimated;And the process of mass data can be applicable to preferably.
3) user can determine its application scenarios according to the assessment result of info web further.Such as, when evaluated info web includes advertisement on line, the input effect of advertisement can be estimated, and the input situation of advertising message is adjusted according to the effect of advertising assessing gained, such as, the release time section of advertisement is adjusted, to improve its clicking rate;Again such as, the targeted customer of advertisement is adjusted, to improve the matching degree etc. of advertisement and user, thus improve the efficiency of advertisement putting and the conversion ratio of advertisement.
Accompanying drawing explanation
By reading the detailed description that non-limiting example is made made with reference to the following drawings, the other features, objects and advantages of the present invention will become more apparent upon:
Fig. 1 is a kind of computer implemented method flow diagram for webpage is estimated according to an aspect of the present invention;
Fig. 2 is a kind of method flow diagram for webpage is estimated in accordance with a preferred embodiment of the present invention;
Fig. 3 is a kind of method flow diagram for webpage is estimated of another preferred embodiment according to the present invention;
Fig. 4 is a kind of method flow diagram for webpage is estimated of another preferred embodiment according to the present invention;
Fig. 5 is the structural representation of a kind of webpage apparatus for evaluating for webpage is estimated according to an aspect of the present invention;
Fig. 6 is the structural representation of a kind of webpage apparatus for evaluating for webpage is estimated according to a preferred embodiment of the present invention;
Fig. 7 is the structural representation of a kind of webpage apparatus for evaluating for webpage is estimated of another preferred embodiment according to the present invention;
Fig. 8 is the structural representation of a kind of webpage apparatus for evaluating for webpage is estimated of another preferred embodiment according to the present invention;
In accompanying drawing, same or analogous accompanying drawing labelling represents same or analogous parts.
Detailed description of the invention
Below in conjunction with accompanying drawing, the present invention is described in further detail.
Fig. 1 illustrates a kind of computer implemented method flow diagram for webpage is estimated according to an aspect of the present invention.The method according to the invention includes step S1, step S2, step S3 and step S4.
Wherein, the method according to the invention is realized by computer equipment.Described computer equipment include a kind of can according to setting in advance or the instruction of storage, automatically carrying out the electronic equipment of numerical computations and/or information processing, its hardware includes but not limited to microprocessor, special IC (ASIC), programmable gate array (FPGA), digital processing unit (DSP), embedded device etc..Described computer equipment includes the network equipment and/or subscriber equipment.Wherein, the described network equipment includes but not limited to server group that single network server, multiple webserver form or the cloud being made up of a large amount of main frames or the webserver based on cloud computing (CloudComputing), wherein, cloud computing is the one of Distributed Calculation, the super virtual machine being made up of a group loosely-coupled computer collection.Described subscriber equipment includes but not limited to any to be carried out the electronic product of man-machine interaction by modes such as keyboard, mouse, remote controller, touch pad or voice-operated devices with user, such as, personal computer, panel computer, smart mobile phone, PDA, game machine or IPTV etc..Wherein, the network residing for described subscriber equipment and the network equipment includes but not limited to the Internet, wide area network, Metropolitan Area Network (MAN), LAN, VPN etc..
It should be noted that; described subscriber equipment, the network equipment and network are only for example; other existing or subscriber equipment, the network equipment and networks of being likely to occur from now on, as being applicable to the present invention, within also should being included in scope, and are incorporated herein with way of reference.
The method according to the invention is used for obtaining webpage assessment models, and by described webpage assessment models, webpage is estimated.
Wherein, heretofore described model includes the model set up based on machine learning mode.It is theoretical based on theory of probability, statistics, Approximation Theory, convextiry analysis, algorithm complex that described machine learning is a class, automatically analyzes and obtain rule from data, and the method that unknown data is predicted by assimilated equations.
The model set up based on described machine learning mode by the training of mass data finds rule therein, and can run based on this rule so that unknown data to be predicted.Further, described model can by optimizing self estimated performance for unknown data based on the training of conventional data.
Wherein, described webpage includes multinomial feature relevant information.Described feature relevant information includes the various information relevant to webpage, it is preferable that described feature relevant information includes the every terms of information being likely to affect the reception degree for the information issued for webpage.
Preferably, described webpage includes for the advertising message to user's release product or information on services, and correspondingly, the feature relevant information of described advertising message includes the every terms of information being likely to affect the effect of advertising of this advertising message.Described webpage assessment models includes the advertisement evaluations model for assessing advertisement.
It is highly preferred that described advertising message can be embedded in other webpages.Such as, the advertising message in news web page it is embedded in the form of the window that suspends;Again such as, it is embedded in the advertising message in webpage with the form such as picture, audio frequency and video, more such as, is presented in the advertising message etc. in webpage with the form such as text, link.
Wherein, in the computer equipment of the method for performing the present invention that is stored in that can prestore according to multiple webpages of the present invention and feature relevant information thereof, or, the computer equipment performing the present invention other equipment connected by it obtain.Preferably, described computer equipment can obtain this webpage and feature relevant information thereof from one or more network equipments of publishing web page.It is highly preferred that described webpage can be issued by this computer equipment self, and by other information relevant to this advertising message via network monitoring or reception, determine the feature relevant information of this webpage.
Such as, after computer equipment issues one or more webpage on predetermined website, in the given time, receive based on predetermined interface and to feed back from each subscriber equipment end, user's operation information to each webpage, and after the information obtained is added up, it is determined that the feature relevant information of each webpage.
Preferably, described feature relevant information includes but not limited to following at least any one information:
1) object-related information promoting object that described webpage comprises.Described object-related information includes but not limited to the every terms of information of the object that webpage promotes.Wherein, the object that described webpage is promoted includes but not limited to various product and service.Preferably, described object-related information includes but not limited to following at least any one:
A) the description information of object is promoted;Such as, the content of copy;Again such as, the picture description information etc. comprised in display advertising.
B) classification information of object is promoted;Described classification information includes but not limited to the classification done in advance for various product or service, for instance, it is the classifications such as electrical equipment, number, article of everyday use by product classification.
C) targeted customer's type information of object is promoted.
D) temporal information is promoted, for instance, the persistent period that webpage is promoted is 2 two months, again such as, 12:00 to the 17:00 that time is every day etc. that web advertisement represents.
2) the clicked information of webpage;Described clicked information includes but not limited to following any one:
A) the frequency information that webpage is clicked;
B) time related information when webpage is clicked;Such as, the time point that advertisement is such as clicked;Again such as, the peak period etc. of advertisement is clicked in one day.
3) user related information of the user of webpage clicking.When the user of webpage clicking is login user, can obtaining the user related information of this user, described user related information includes but not limited to following at least any one:
A) subscriber information message, for instance age, occupation, individual's description information etc.;
B) information-setting by user, for instance the advertisement type information that user setup receives, realm information etc. that user is interested;
C) history information of user;Such as, the advertising message etc. that user clicked.
D) other relevant informations, it include computer equipment obtainable other with user-dependent information, for instance, the regional information etc. that the information of terminal unit that user uses, user are presently in.
4) webpage represent type information.Preferably, what described webpage represented that type information includes advertisement represents type information, and wherein, described showing advertisement type information includes but not limited to following any one:
A) picture/mb-type;
B) literal type;
C) audio types;
D) video type.
It is highly preferred that when described webpage includes advertising message, described in represent type information also include advertisement clicked time status information;Such as, for the advertisement of picture/mb-type, it is that thumbnail state still dynamically represents state when it is clicked;Again such as, for audio frequency and video series advertisements, it is broadcast state still non-broadcast state etc. when it is clicked.
Preferably, when described webpage includes advertising message, described feature relevant information also includes the webpage relevant information of webpage belonging to advertising message.Wherein, described webpage relevant information includes but not limited to affect the relevant information self representing and clicking.Preferably, described webpage relevant information includes but not limited to following at least any one:
A) the type of webpage information of webpage belonging to advertisement;
B) page layout information;
C) webpage targeted customer type information etc..
Being described in detail below with reference to Fig. 1, in step sl, computer equipment obtains one group of target sample group of webpage assessment models to be set up and corresponding least one set sample for reference group.Wherein, the least one set sample for reference group of described target sample group and correspondence thereof includes multiple advertising message respectively.
Wherein, described target sample group and corresponding least one set sample for reference group can prestore and be stored in this computer equipment, or, this computer equipment other equipment connected by it obtain.
Such as, the network equipment that computer equipment is connected by self obtains a web pages of this network equipment pre-stored as target sample group, and the another two network equipment connected by self obtains multiple webpages of this another two network equipment pre-stored respectively, using as two the sample for reference groups corresponding with this target sample group.
Preferably, the least one set sample for reference group of described target sample group and correspondence thereof comprises the webpage of magnanimity respectively.
It is highly preferred that the least one set sample for reference group of described target sample group and correspondence thereof comprises the advertising message of magnanimity respectively.
Specifically, the mode of one group of target sample group and corresponding least one set sample for reference group that described computer equipment obtains webpage assessment models to be set up include but not limited to following any one:
1) it is stored in this first acquisition device 1 said computer equipment when the sample for reference group of the target sample group divided in advance and correspondence thereof prestores, or, prestoring when being stored in other equipment being associated with this first acquisition device 1 said computer equipment, computer equipment directly obtains preselected target sample group and corresponding sample for reference group.
2) according to the one or more feature relevant information of webpage, multiple webpages are divided into multiple sample group;Then, by the many web pages divided selecting least one set as described target sample group, and by other web pages remaining selecting one or more groups as the sample for reference group corresponding with this target sample group.
Specifically, described computer equipment according to the one or more feature relevant information of webpage multiple advertising messages are divided into the mode of multiple sample group include but not limited to following any one:
Multiple webpages are divided into multiple sample group by a) the division condition according to pre-determining;Its
In, described division condition is determined according to one or more feature relevant information.
Such as, current multiple advertising messages are divided into multiple sample group by the time period that represents according to advertising message.Again such as, first the type that represents according to advertising message carries out Preliminary division, further according to click advertising message the occupational information of user advertising message is made Further Division, to obtain multiple sample group.
B) according to one or more feature relevant information multiple webpages are performed cluster operation, and multiple groups cluster obtained as sample group.
Wherein, cluster operation is for being divided into different groups similar object so that the member object in same group all has similar attribute.It will be understood by those skilled in the art that various cluster operation method, such as based on the structural clustering method of range measurement, k-means clustering method, QT clustering method etc., as being applicable to the present invention, should be included in protection scope of the present invention.
After having divided multiple sample group, this many web pages divided select least one set as the mode of described target sample group include but not limited to following any one:
1) first predetermined training method is adopted to be trained obtaining initial training model to each group, and the effect according to initial training model, determine target sample group.
Such as, for acquired n sample group, adopt predetermined support vector machine training method that the advertising message in this n group sample group is trained, to obtain n initial training model respectively, the quality of this n initial training model is marked by computer equipment, and select cam group corresponding to initial training model wherein marking minimum, it can be used as target sample group.
2) by user according to self-demand, by the multiple sample groups divided select target sample group.
3) to each sample group obtained, select each sample group as target sample group one by one.
Then, computer equipment is by selecting one or more groups as the sample for reference group corresponding with this target sample group in other web pages remaining.
Specifically, described by other web pages remaining selecting one or more groups mode as the sample for reference group corresponding with this target sample group include but not limited to any one mode following:
1) computer equipment is by other web pages remaining, and one or more groups of the satisfied screening conditions subscribed of selection is as sample for reference group.
Such as, advertising message comprises the information of clicked time point, and predetermined filtering condition is the advertising message clicked time belonging to time period " 22:00~0:00 ", then computer equipment is determined, when in sample group, the clicked time point information of each advertising message belonged in this time period, this sample group is sample for reference group.
2) computer equipment is by randomly choosing one or more groups sample for reference group as this target sample group in other web pages remaining.
3) similarity between one or more groups and this target sample group in other web pages remaining determined by computer equipment;Then, computer equipment is according to described similarity, and selects rule based on predetermined, by selecting the least one set webpage similar to described target sample group in one or more groups webpage described, as the sample for reference group of described target sample group.
Wherein, the multinomial feature relevant information that computer equipment can comprise according to webpage, calculate the similarity between each other sample group and target sample groups from multiple dimensions.
Wherein, the described predetermined sample for reference group selecting rule to be used for selecting described target sample group according to similarity.
Such as, when similarity adopts percentage ratio to represent, predetermined select rule can include " using other group advertising messages more than 80% of similarity as sample for reference group ";Again such as, pre-defined rule can include " using 10 groups of advertising messages before most like with target sample group as sample for reference group " etc..
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, the implementation of any one group of target sample group obtaining webpage assessment models to be set up and corresponding least one set sample for reference group, should be included in the scope of the present invention.
Then, in step s 2, computer equipment is according to described target sample group and described least one set sample for reference group, it is thus achieved that in described least one set sample for reference group part or all refer to the sample group weight adjustment information relative to described target sample group.
Wherein, described weight adjustment information include but not limited to following any one:
1) packet weight information;It includes each sample for reference group comprising multiple advertising message weight information relative to target sample group.
2) Model Weight information;It includes reference group model that each sample for reference group set up by the predetermined training method weight information relative to target sample group.
3) difference weight model information;It includes the model for determining the weight difference that target group model is adjusted.
Specifically, the embodiment corresponding with described three kinds of weight adjustment information by follow-up with reference to the embodiment of Fig. 2 to Fig. 4 in described in detail, repeat no more herein.
Then, in step s3, the computer equipment described weight adjustment information corresponding to described target sample group, each sample for reference group and described predetermined training method, perform migration operation, it is thus achieved that the webpage assessment models of described target sample group.
Wherein, described predetermined training method includes but not limited to the predetermined various machine learning modes that can be used for setting up assessment models.Preferably, described predetermined training method includes but not limited to any one mode following:
1) support vector machine training method;
2) neural metwork training mode;
3) linear regression training method;
4) logistic regression training method;
5) decision tree training method.
Specifically, the computer equipment described weight adjustment information corresponding to described target sample group, each sample for reference group and described predetermined training method, perform migration operation, obtain described target sample group webpage assessment models detailed description of the invention by follow-up with reference to the embodiment of Fig. 2 to Fig. 4 in described in detail, repeat no more herein.
Then, in step s 4, one or more webpages to be assessed are estimated by computer equipment based on described webpage assessment models.
Preferably, described step S4 farther includes step S41 (not shown), step S42 (not shown) and step S43 (not shown).
In step S41, computer equipment obtains the feature relevant information of webpage to be assessed.
Specifically, obtain the mode of the feature relevant information of webpage to be assessed include but not limited to following any one:
1) multiple webpages and the feature relevant information thereof of self pre-stored are directly obtained;
2) obtained from other equipment that it connects by described computer equipment.
Preferably, described computer equipment can obtain this advertising message and feature relevant information thereof from one or more network equipments of publishing web page.
It is highly preferred that described webpage can be issued by this computer equipment self, and by other information relevant to this webpage via network monitoring or reception, determine the feature relevant information of this webpage.
Then, in step S42, computer equipment is according to the feature relevant information obtained, it is determined that this target sample group corresponding to webpage to be assessed.
Wherein, described computer equipment is according to the feature relevant information obtained, determine the mode of this target sample group corresponding to webpage to be assessed and the implementation 2 of abovementioned steps S1) described in the computer equipment mode that multiple webpages is divided into multiple sample group according to the one or more feature relevant information of webpage same or similar, do not repeat them here.
Then, in step S43, this webpage to be evaluated is performed evaluation operation based on the webpage assessment models corresponding with this sample group of target by computer equipment.
The described assessment result that webpage to be assessed performs evaluation operation is used for the reception degree of the information indicating user to issue for webpage.Preferably, described assessment result includes following at least any one information:
1) probabilistic information that webpage is clicked;
2) the matching degree information of the user of webpage and predefined type.
Preferably, when described webpage to be assessed includes advertising message to be assessed, described Evaluated effect includes effect of advertising, the degree that described effect of advertising are accepted by user for indicating advertising message.It is highly preferred that described effect of advertising include but not limited to following at least any one:
1) probabilistic information that advertising message is clicked.
2) the matching degree information of the user of advertising message and predefined type.Wherein, described predefined type includes the classification after user is divided by the one or more user related information according to user.
Such as, the classification obtained after user being divided according to user's occupational information, wherein, described user's occupational information can be obtained by the data of family;Again such as, the classification etc. obtained after user being divided according to the age of user.
Wherein, described matching degree information for indicate advertising message and this predefined type user demand between agree with degree.Wherein, computer equipment can determine described matching degree information by following at least any of mode:
A) determine according to the description information of advertising message and the goodness of fit of user related information
Degree of joining information;Such as, in the text message corresponding to audience and subscriber data from
I describes the characters matching rate of partial content to determine matching degree information.
B) the matching degree information determined according to the conversion ratio of advertising message.Such as, directly using the conversion ratio of the advertising message matching degree information as this advertisement;Again such as, the conversion ratio relative to different types of user of advertising message is done the result after normalized as matching degree information etc..
Such as, computer equipment obtains following three feature relevant informations of advertisement Ad1 to be assessed in step S41:
Targeted customer's type of webpage belonging to advertising message: be positioned at the user of Area1, Area2 and Area3;
The clicked number of times of advertising message: 2324 times;
User's occupation of advertising message: company clerk.
Further, computer equipment in step S42 according to above-mentioned three information, it is determined that the target cam group corresponding to advertisement Ad1 to be assessed is Ad_group1.
Then computer equipment determines that the advertisement evaluations model corresponding to this target sample group Ad_group1 is Ad_Model1, wherein, the advertisement evaluations model Ad_Mode1 that this has been set up is for targeted customer's type information of webpage belonging to advertisement, the clicked number of times of advertising message, and click the user related information of the user of advertisement, the matching degree with the user being positioned at area Area1 of assessment advertising message, then perform device and adopt this initial assessment model Ad_Mode1, these three feature relevant informations execution evaluation operation based on the advertisement Ad1 to be evaluated that the second acquisition device obtains, obtain this advertisement Ad1 to be evaluated relative to the matching degree information of user being positioned at area Area1.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, any implementation that this webpage to be assessed performs evaluation operation based on the webpage assessment models corresponding with this sample group of target, should be included in the scope of the present invention.
The method according to the invention, by performing Data Migration or model migration operation, based on current existing data or model, corresponding with target sample group, rational assessment models can be set up, to be estimated each advertising message belonging to target sample group determining its effect of advertising.Further, the method according to the invention, it is possible to quickly realize convergence, it is established that stable assessment models, therefore can be applicable to the process of mass data preferably.Additionally, user can adjust the input situation of advertising message further according to effect of advertising, for instance, adjust the release time section of advertisement, to improve its clicking rate;Again such as, the targeted customer of advertisement is adjusted, to improve the matching degree etc. of advertisement and user, thus improve the efficiency of advertisement putting and the conversion ratio of advertisement.
Fig. 2 illustrates a kind of method flow diagram for webpage is estimated in accordance with a preferred embodiment of the present invention.Method according to the present embodiment, described weight adjustment information includes each sample for reference group in described least one set sample for reference group and is respectively relative to the packet weight information of described target sample group.Method according to the present embodiment includes step S1, step S2, step S3 and step S4.Wherein, described step S2 farther includes step S21, and described step S3 farther includes step S31 and step S32.
Wherein, step S1, step S4 are described in detail in reference to the embodiment shown in Fig. 1, and are incorporated herein by reference, repeat no more.
Preferably, described packet weight information includes each sample for reference group the first similarity relative to target sample group.
In the step s 21, when recanalization information of holding power includes that in described least one set sample for reference group, each sample for reference group is respectively relative to the packet weight information of described target sample group, each feature relevant information to each sample for reference group, obtain this stack features relevant information the first similarity relative to this feature relevant information of target sample group, to determine each reference group packet weight information relative to target sample group according to this first similarity.
Such as, each advertising message organized in sample for reference group and target sample group all comprises two feature relevant informations: user's occupational information and advertisement positional information in webpage.Further, the predetermined accounting weighted value of user's occupational information is 20%, and the predetermined accounting weight of the positional information in advertisement webpage belonging to it is 80%.Then computer equipment first determines the often group sample for reference group value relative to the similarity Sim1 of target sample group according to user's occupational information, and further according to advertisement positional information in webpage, this determines the often group sample for reference group value relative to the similarity Sim2 of target sample group;Finally determine that often group sample for reference group is relative to first similarity (Sim1*20%+Sim2*80%) of target sample group, and the value obtained is made the packet weight information of this group sample for reference group.
In step S31, the computer equipment packet weight information according to each sample for reference group, the feature relevant information of each webpage in each sample for reference group is adjusted, to obtain the feature relevant information of each webpage after adjusting.
The first example according to the present invention, target sample group Group1 has m corresponding sample for reference group Group1_1, Group1_2 ..., Group1_m, and, this m sample for reference group is relative to the packet weight information respectively W of this target sample group1, W2, W3..., Wm, wherein, the packet weight information of the sample for reference group of i-th group is Wi, wherein, i is natural number, and 1≤i≤m.The then computer equipment packet weight according to each sample for reference group, the feature relevant information of each advertising message in this sample for reference group is adjusted, to obtain each group Group1_1 ', the Group1_2 of the feature relevant information after comprising adjustment ', ..., Group1_m '.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, any packet weight information according to each sample for reference group, the feature relevant information of each webpage in each sample for reference group is adjusted, to obtain the implementation of the feature relevant information of each webpage after adjusting, should be included in the scope of the present invention.
Then, in step s 32, computer equipment adopts predetermined training method to be trained based on the feature relevant information after the adjustment of described target sample group and described sample for reference group, with the webpage assessment models that the webpage obtained with described target sample group adapts.
Continue first example of the present invention is illustrated, computer equipment is based on each group Group_1, Group1_1 of obtaining ', Group1_2 ', ..., the feature relevant information of all advertising messages comprised in Group1_m ', predetermined support vector machine training method is adopted to be trained, to obtain the advertisement evaluations model corresponding with target sample group Group1.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, any feature relevant information based on after the adjustment of described target sample group and described sample for reference group adopts predetermined training method to be trained, with the implementation of the webpage assessment models that the webpage obtained with described target sample group adapts, should be included in the scope of the present invention.
Method according to the present embodiment, by using the Data Migration of reference group as target group data, achieve the extensive of target group data to a certain extent so that the model set up is when less relative to the deviation of former target group data, and the variation when it is extensive is also less.So that the webpage assessment models obtained has good effect when assessment and target sample group have other advertising messages of similar characteristic.Further, data need not be carried out iteration repeatedly according to the method for the present embodiment, so processing mass data, for instance convergence can be rapidly achieved during the process of the advertising message of magnanimity, greatly improve the operational efficiency of computer equipment, there is greater advantage.
Fig. 3 illustrates a kind of method flow diagram for webpage is estimated according to another preferred embodiment of the present invention.Weight adjustment information according to the present embodiment includes Model Weight information.Method according to the present embodiment includes step S1, step S2, step S3 and step S4.Wherein, described step S2 farther includes step S22 and step S23, and described step S3 farther includes step S33.
Wherein, step S1 is described in detail in reference to the embodiment shown in Fig. 1, and is incorporated herein by reference, repeats no more.
In step S22, computer equipment obtains and described target sample group and each sample for reference group initial assessment model corresponding, that obtain based on predetermined training method respectively.
Specifically, described acquisition and described target sample group and each sample for reference group mode corresponding, initial assessment model that is that obtain based on predetermined training method respectively include but not limited to following any one:
1) the pre-established initial assessment model not corresponding with each sample components is obtained.
Specifically, the mode of the initial assessment model not corresponding with each sample components that described acquisition is pre-established include but not limited to following any one:
A) this of computer equipment self pre-stored is directly obtained pre-established with each sample group
Each corresponding respectively initial assessment model.
B) by other equipment being connected with computer equipment obtain this pre-established and each
Each initial assessment model that sample components is not corresponding.
2) predetermined training method is adopted to be trained each sample group, to obtain the initial assessment model not corresponding with each sample components.
The second example according to the present invention, target sample group Group2 has the sample for reference group Group2_1 that p group is corresponding, Group2_2, ..., Group2_p, then this p group sample group is adopted preselected neural metwork training mode to be trained by computer equipment respectively, to obtain initial assessment model M odel2_1 corresponding with each sample for reference group respectively, Model2_2 ..., Model2_p.Similarly, target sample group Group2 is adopted preselected nerve net training method to be trained by computer equipment, it is thus achieved that corresponding initial assessment model M odel2.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, any acquisition and described target sample group and each sample for reference group implementation corresponding, initial assessment model that is that obtain based on predetermined training method respectively, should be included in the scope of the present invention.
In step S23, computer equipment obtains the Model Weight information that the initial assessment model of each sample for reference group is respectively relative to the webpage of described target sample group.
Specifically, for each initial assessment model, computer equipment adopts initial assessment model that the webpage in target sample group performs evaluation operation the accuracy according to assessment result, determines this initial assessment model Model Weight information relative to this target sample group.
Wherein, the accuracy of described assessment result can by obtaining after being compared by the actual information of assessment result Yu webpage.
Preferably, when described webpage includes advertising message, computer equipment adopts initial assessment model that the advertising message in target sample group is performed evaluation operation to obtain the effect of advertising of each advertising message, and the accuracy according to the effect of advertising obtained, determine this initial assessment model Model Weight information relative to this target sample group.
Wherein, the accuracy of the assessment result of described advertising message can be obtained after being compared by the actual advertisement effect of effect of advertising assessment obtained and advertising message.
Such as, the effect of advertising of assessment gained of each advertising message in target sample group are compared by initial assessment model with the effect of advertising of its reality, have in evaluated obtained effect of advertising 80% with the difference of its actual effect of advertising in range of allowable error time, it is determined that the accuracy rate of this initial assessment model is 80%.
Preferably, described computer equipment can adopt similar method, it is thus achieved that the Model Weight information of target sample group self.
Preferably, computer equipment can directly using described accuracy as described Model Weight value.It is highly preferred that described Model Weight value can be pre-stored in this computer equipment.It is highly preferred that described Model Weight value can be pre-stored in other equipment that this computer equipment is connected.
Continue the second example is illustrated, computer equipment is respectively adopted p initial assessment model M odel2_1, Model2_2, ..., advertising message in target sample group Group2 is estimated by Model2_p, with obtain the multiple advertising messages in Group2 be based respectively on this each initial assessment model obtain effect of advertising, and by by assess gained effect of advertising compared with actual advertisement effect, to obtain the accuracy Ac of described p initial assessment model respectively1, Ac2..., Acp, then computer equipment is using this each accuracy as the Model Weight information of the initial assessment model of its correspondence.Similarly, computer equipment adopts the initial assessment model M odel2 of target sample group Group2 that the advertising message in target sample group Group2 is estimated, it is based respectively on this initial assessment model M odel2 effect of advertising obtained obtaining the multiple advertising messages in target sample group Group2, and by the effect of advertising by assessing gained compared with actual advertisement effect, thus obtaining the Model Weight value Ac of this initial assessment model M odel2g。
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, the initial assessment model of each sample for reference group of any acquisition is respectively relative to the implementation of the Model Weight information of the webpage of described target sample group, should be included in the scope of the present invention.
In step S33, the initial assessment model of described target group, according to the initial assessment model of each sample for reference group described and Model Weight information thereof, is adjusted by computer equipment, to obtain the webpage assessment models adapted with described target sample group.
Preferably, computer equipment is according to the initial assessment model of each sample for reference group described and Model Weight information thereof, using the weighted mean of the initial assessment model of each sample for reference group described as described webpage assessment models.
Continuing the second example is illustrated, the weighted mean that advertisement evaluations model is each initial assessment model corresponding for Group2 determined by computer equipment, that is the final advertisement evaluations model M odel2 ' obtained including: (Model2*Acg+Model2_1*Ac1+Model2_2*Ac2+...+Model2_p*Acp)/(Acg+Ac1+Ac2+...+Acp)。
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, the initial assessment model of each sample for reference group described in any basis and Model Weight information thereof, the initial assessment model of described target group is adjusted, to obtain the implementation of the webpage assessment models adapted with described target sample group, should be included in the scope of the present invention.
Preferred version according to the present embodiment, computer equipment obtains the initial assessment model corresponding with target sample group, and based on the initial assessment model that all webpages of described target sample group and each sample for reference group generate, and obtain the Model Weight information of these two initial assessment models respectively, to obtain the final webpage assessment models corresponding with target sample group.
Method according to the present embodiment, realizes less deviation by the initial assessment model corresponding with target sample group, and by the initial assessment model of other sample groups after adjusting in conjunction with weight, decreases the variation of model evaluation result.So that the final webpage assessment models obtained has the suitability widely and assessment result more accurately.Further, according to the migration operation of the present embodiment, each data need not being iterated operation, so convergence can be rapidly achieved, therefore, it is possible to be well applicable to the process of mass data, the assessment for the advertising message of current magnanimity can obtain good effect.
Fig. 4 illustrates a kind of method flow diagram for webpage is estimated according to another preferred embodiment of the present invention.Method according to the present embodiment, described weight adjustment information includes difference weight model.Method according to the present embodiment includes step S1, step S2, step S3 and step S4.Wherein, described step S2 farther includes step S24, and described step S3 farther includes step S34.
Wherein, step S1 and step S4 is described in detail in reference to the embodiment shown in Fig. 1, and is incorporated herein by reference, repeats no more.
In step s 24, computer equipment is according to described target sample group, described least one set sample for reference group and described predetermined training method, it is thus achieved that for performing the difference weight model of migration operation.
Specifically, according to described target sample group, described least one set sample for reference group and described predetermined training method, it is thus achieved that the mode of difference weight model for performing migration operation include but not limited to following any one:
1) computer equipment obtains the pre-established difference weight model corresponding with each sample group and predetermined training method.
Specifically, the mode of the described pre-established difference weight model corresponding with each sample group and predetermined training method include but not limited to following any one:
A) computer equipment self pre-stored, this pre-established difference weight model corresponding with each sample group and predetermined training method are directly obtained.
B) by other equipment being connected with computer equipment obtain this pre-established difference weight model corresponding with each sample group and predetermined training method.
2) computer equipment first adopts described predetermined training method that the webpage of the least one set sample for reference group of described target sample group and correspondence thereof is trained, to obtain pre-evaluation model;Then, described pre-evaluation model is adopted to continue executing with training described target sample group, to obtain difference evaluation model;Then computer equipment determines difference weight model according to described pre-evaluation model and described difference evaluation model.
nullThe 3rd example according to the present invention,Computer equipment adopts the predetermined neural metwork training mode q group sample for reference group Group3_1 to target sample group Group3 and correspondence thereof,Group3_2,...,Whole advertising messages that Group3_q comprises are trained,To obtain the pre-evaluation model W_Model based on all advertising messages,Then,Computer equipment individually adopts the Group3 advertising message comprised that this pre-evaluation model W_Model proceeds training,To obtain difference evaluation model W_Model ',Then computer equipment is according to the pre-evaluation model W_Model obtained and difference evaluation model W_Model ',Determine that the difference weight model corresponding with target sample group Group3 is | W_Model-W_Model ' |.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, any according to described target sample group, described least one set sample for reference group and described predetermined training method, obtain the implementation being used for performing the difference weight model of migration operation, should be included in the scope of the present invention.
In step S34, computer equipment adopts described predetermined training method that the webpage of described target sample group is trained, to obtain initial assessment model.
Continuing the 3rd example is illustrated, computer equipment adopts predetermined neural metwork training mode that target sample group Group3 is trained, to obtain the initial assessment model M odel3 corresponding with target sample group Group3.
In step s 35, described initial assessment model, according to described pre-evaluation model and described difference weight model, is adjusted by computer equipment, with the webpage assessment models that the webpage obtained with this target sample group adapts.
Continue the 3rd example is illustrated, computer equipment is according to determined difference weight model | W_Model-W_Model ' | in step s 24, adjust the initial assessment model M odel3 corresponding with target sample group Group3 obtained in step S34, to obtain final advertisement evaluations model M odel3 '=Model3+ | the W_Model-W_Model ' | after adjusting.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, the initial assessment model of each sample for reference group described in any basis and Model Weight information thereof, the initial assessment model of described target group is adjusted, to obtain the implementation of the webpage assessment models adapted with described target sample group, should be included in the scope of the present invention.
Method according to the present embodiment, the webpage assessment models obtained contains the initial assessment model corresponding with target sample group, current model is made to have less deviation, and, advertising message for current goal sample group, current initial assessment model is adjusted so that initial assessment model has better adaptability, thus being not susceptible to variation by difference weight model.Additionally, the method according to the present embodiment can be rapidly achieved convergence when processing mass data, alleviate the burden of computer equipment, decrease required computer equipment quantity.
Fig. 5 illustrates the structural representation of a kind of webpage apparatus for evaluating for webpage is estimated according to an aspect of the present invention.Webpage apparatus for evaluating according to the present invention includes the first acquisition device 1, Weight Acquisition device 2, model acquisition device 3 and evaluation operation device 4.
Wherein, it is used for obtaining webpage assessment models according to the webpage apparatus for evaluating of the present invention, and by described webpage assessment models, webpage is estimated.
Wherein, described webpage includes multinomial feature relevant information.Described feature relevant information includes the various information relevant to webpage, it is preferable that described feature relevant information includes the every terms of information being likely to affect the reception degree for the information issued for webpage.
Preferably, described webpage includes for the advertising message to user's release product or information on services, and correspondingly, the feature relevant information of described advertising message includes the every terms of information being likely to affect the effect of advertising of this advertising message.Described webpage assessment models includes the advertisement evaluations model for assessing advertisement.
It is highly preferred that described advertising message can be embedded in other webpages.Such as, the advertising message in news web page it is embedded in the form of the window that suspends;Again such as, it is embedded in the advertising message in webpage with the form such as picture, audio frequency and video, more such as, is presented in the advertising message etc. in webpage with the form such as text, link.
Wherein, can prestore according to multiple webpages of the present invention and feature relevant information thereof in the webpage apparatus for evaluating being stored in for performing the present invention, or, the webpage apparatus for evaluating performing the present invention other equipment connected by it obtain.Preferably, described webpage apparatus for evaluating can obtain this webpage and feature relevant information thereof from one or more network equipments of publishing web page.It is highly preferred that described webpage can be issued by this webpage apparatus for evaluating self, and by other information relevant to this webpage via network monitoring or reception, determine the feature relevant information of this webpage.
Such as, after advertisement evaluations device issues one or more webpage on predetermined website, in the given time, receive based on predetermined interface and to feed back from each subscriber equipment end, user's operation information to each webpage, and after the information obtained is added up, it is determined that the feature relevant information of each webpage.
Preferably, described feature relevant information includes but not limited to following at least any one information:
1) object-related information promoting object that described webpage comprises.Described object-related information includes but not limited to the every terms of information of the object that webpage promotes.Wherein, the object that described webpage is promoted includes but not limited to various product and service.Preferably, described object-related information includes but not limited to following at least any one:
A) the description information of object is promoted;Such as, the content of copy;Again such as, the picture description information etc. comprised in display advertising.
B) classification information of object is promoted;Described classification information includes but not limited to the classification done in advance for various product or service, for instance, it is the classifications such as electrical equipment, number, article of everyday use by product classification.
C) targeted customer's type information of object is promoted.
D) temporal information is promoted, for instance, the persistent period that webpage is promoted is 2 two months, again such as, 12:00 to the 17:00 that time is every day etc. that web advertisement represents.
2) the clicked information of webpage;Described clicked information includes but not limited to following any one:
A) the frequency information that webpage is clicked;
B) time related information when webpage is clicked;Such as, the time point that advertisement is such as clicked;Again such as, the peak period etc. of advertisement is clicked in one day.
3) user related information of the user of webpage clicking.When the user of webpage clicking is login user, can obtaining the user related information of this user, described user related information includes but not limited to following at least any one:
A) subscriber information message, for instance age, occupation, individual's description information etc.;
B) information-setting by user, for instance the advertisement type information that receive of user setup, the realm information etc. that user is interested;
C) history information of user;Such as, the advertising message etc. that user clicked.
D) other relevant informations, it include webpage apparatus for evaluating obtainable other with user-dependent information, for instance, the regional information etc. that the information of terminal unit that user uses, user are presently in.
4) webpage represent type information.Preferably, what described webpage represented that type information includes advertisement represents type information, and wherein, described showing advertisement type information includes but not limited to following any one:
A) picture/mb-type;
B) literal type;
C) audio types;
D) video type.
It is highly preferred that when described webpage includes advertising message, described in represent type information also include advertisement clicked time status information;Such as, for the advertisement of picture/mb-type, it is that thumbnail state still dynamically represents state when it is clicked;Again such as, for audio frequency and video series advertisements, it is broadcast state still non-broadcast state etc. when it is clicked.
Preferably, when described webpage includes advertising message, described feature relevant information also includes the webpage relevant information of webpage belonging to advertising message.Wherein, described webpage relevant information includes but not limited to affect the relevant information self representing and clicking.Preferably, described webpage relevant information includes but not limited to following at least any one:
A) the type of webpage information of webpage belonging to advertisement;
B) page layout information;
C) webpage targeted customer type information etc..
Wherein, the degree received by user for indicating advertising message according to effect of advertising of the present invention.Preferably, described effect of advertising include but not limited to following at least any one:
1) probabilistic information that advertising message is clicked.
2) the matching degree information of the user of advertising message and predefined type.Wherein, described predefined type includes the classification after user is divided by the one or more user related information according to user.
Such as, the classification obtained after user being divided according to user's occupation;Again such as, the classification etc. obtained after user being divided according to the age of user.
Wherein, described matching degree information for indicate advertising message and this predefined type user demand between suit degree.Wherein, computer equipment can determine described matching degree information by following at least any of mode:
A) matching degree information is determined according to the description information of advertising message and the goodness of fit of user related information;Such as, the text message corresponding to audience determines matching degree information with the characters matching rate of self portrait partial content in subscriber data.
B) the matching degree information determined according to the conversion ratio of advertising message.Such as, directly using the conversion ratio of the advertising message matching degree information as this advertisement;Again such as, the conversion ratio relative to different types of user of advertising message is done the result after normalized as matching degree information etc..
Being described in detail below with reference to Fig. 5, the first acquisition device 1 obtains one group of target sample group of webpage assessment models to be set up and corresponding least one set sample for reference group.Wherein, the least one set sample for reference group of described target sample group and correspondence thereof includes multiple advertising message respectively.
Wherein, described target sample group and corresponding least one set sample for reference group can prestore and be stored in this webpage apparatus for evaluating, or, other equipment that the computer equipment belonging to this webpage apparatus for evaluating connects obtain.
Such as, the network equipment that first acquisition device 1 is connected by self obtains a web pages of this network equipment pre-stored as target sample group, and the another two network equipment connected by self obtains multiple webpages of this another two network equipment pre-stored respectively, using as two the sample for reference groups corresponding with this target sample group.
Preferably, the least one set sample for reference group of described target sample group and correspondence thereof comprises the webpage of magnanimity respectively.
It is highly preferred that the least one set sample for reference group of described target sample group and correspondence thereof comprises the advertising message of magnanimity respectively.
Specifically, the mode of one group of target sample group and corresponding least one set sample for reference group that described first acquisition device 1 obtains webpage assessment models to be set up include but not limited to following any one:
1) it is stored in this first acquisition device 1 said computer equipment when the sample for reference group of the target sample group divided in advance and correspondence thereof prestores, or, prestoring when being stored in other equipment being associated with this first acquisition device 1 said computer equipment, the first acquisition device 1 directly obtains this target sample group and corresponding sample for reference group of having divided in advance.
2) multiple webpages are divided into multiple sample group according to the one or more feature relevant information of webpage by device (not shown) that what the first acquisition device 1 comprised divide;Then, the first selection device (not shown) that first acquisition device 1 comprises is by selecting least one set as described target sample group in the many web pages divided, and, select each target sample group selected by device to first, the second selection device (not shown) that the first acquisition device 1 comprises is by selecting one or more groups as the sample for reference group corresponding with this target sample group in other web pages remaining.
Specifically, described division device according to the one or more feature relevant information of webpage multiple webpages are divided into the mode of multiple sample group include but not limited to following any one:
A) divide the device division condition according to pre-determining, multiple webpages are divided into multiple sample group;Wherein, described division condition is determined according to one or more feature relevant information.
Such as, divide device, according to the time period that represents of advertising message, current multiple advertising messages are divided into multiple sample group.Again such as, divide the device first type that represents according to advertising message and carry out Preliminary division, further according to click advertising message the occupational information of user advertising message is made Further Division, to obtain multiple sample group.
B) according to one or more feature relevant information multiple webpages are performed cluster operation, and multiple groups cluster obtained as sample group.
Wherein, cluster operation is for being divided into different groups similar object so that the member object in same group all has similar attribute.It will be understood by those skilled in the art that various cluster operation method, such as based on the structural clustering method of range measurement, k-means clustering method, QT clustering method etc., as being applicable to the present invention, should be included in protection scope of the present invention.
After dividing device and having divided multiple sample group, device is selected to perform operation by first.Wherein, first select many web pages of being divided by this of device select least one set as the mode of described target sample group include but not limited to following any one:
1) first device is selected first to adopt predetermined training method to be trained obtaining initial training model to each group, and the effect according to initial training model, determine target sample group.
Such as, for acquired n sample group, first selects device to adopt predetermined support vector machine training method that the advertising message in this n group sample group is trained, to obtain n initial training model respectively, first selects device that the quality of this n initial training model is marked, and select cam group corresponding to initial training model wherein marking minimum, it can be used as target sample group.
2) device demand according to user is selected by first, by the multiple sample groups divided select target sample group.
3) first select device each sample group to obtaining, select each sample group as target sample group one by one.
Then, selecting each target sample group that device selects to first, second selects device to be selected one or more groups as the sample for reference group corresponding with this target sample group by other group advertising messages remaining.
Specifically, described second select device by other web pages remaining selecting one or more groups mode as the sample for reference group corresponding with this target sample group include but not limited to any one mode following:
1) second selecting device by other web pages remaining, one or more groups of the satisfied screening conditions subscribed of selection is as sample for reference group.
Such as; advertising message protects the information of clicked time point; and predetermined filtering condition is the advertising message clicked time belonging to time period " 22:00~0:00 "; then second device is selected to determine; when in sample group, the clicked time point information of each advertising message belonged in this time period, this sample group is sample for reference group.
2) second select device by other web pages remaining randomly choose one or more groups sample for reference group as this target sample group.
3) second the first the second similarity determining between one or more groups and this target sample group that device (not shown) is determined in other web pages remaining that device comprises is selected;Then, second selects the son that device comprises to select device (not shown) according to described second similarity, and select rule based on predetermined, by one or more groups webpage described selects the least one set webpage similar to described target sample group, as the sample for reference group of described target sample group.
Wherein, first determines the multinomial feature relevant information that device can comprise according to webpage, calculates the second similarity between each other sample group and target sample groups from multiple dimensions.
Wherein, the described predetermined sample for reference group selecting rule to be used for selecting described target sample group according to the second similarity.
Such as, when the second similarity adopts percentage ratio to represent, predetermined select rule can include " using other group advertising messages more than 80% of similarity as sample for reference group ";Again such as, pre-defined rule can include " using 10 groups of advertising messages before most like with target sample group as sample for reference group " etc..
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, the implementation of any one group of target sample group obtaining webpage assessment models to be set up and corresponding least one set sample for reference group, should be included in the scope of the present invention.
Then, Weight Acquisition device 2 is according to described target sample group and described least one set sample for reference group, it is thus achieved that in described least one set sample for reference group part or all refer to the sample group weight adjustment information relative to described target sample group.
Wherein, described weight adjustment information include but not limited to following any one:
1) packet weight information;It includes each sample for reference group comprising multiple advertising message weight information relative to target sample group.
2) Model Weight information;It includes reference group model that each sample for reference group set up by the predetermined training method weight information relative to target sample group.
3) difference weight model information;It includes the model for determining the weight difference that target group model is adjusted.
Specifically, the embodiment corresponding with described three kinds of weight adjustment information by follow-up with reference to the embodiment of Fig. 6 to Fig. 8 in described in detail, repeat no more herein.
Then, the model acquisition device 3 described weight adjustment information corresponding to described target sample group, each sample for reference group and described predetermined training method, perform migration operation, it is thus achieved that the webpage assessment models of described target sample group.
Wherein, described predetermined training method includes but not limited to the predetermined various machine learning modes that can be used for setting up assessment models.Preferably, described predetermined training method includes but not limited to any one mode following:
1) support vector machine training method;
2) neural metwork training mode;
3) linear regression training method;
4) logistic regression training method;
5) decision tree training method.
Specifically, the model acquisition device 3 described weight adjustment information corresponding to described target sample group, each sample for reference group and described predetermined training method, perform migration operation, obtain described target sample group webpage assessment models detailed description of the invention by follow-up with reference to the embodiment of Fig. 6 to Fig. 8 in described in detail, repeat no more herein.
Then, one or more webpages to be assessed are estimated by evaluation operation device 4 based on described webpage assessment models.
Preferably, described evaluation operation device 4 include the second acquisition device (not shown), second determine device (not shown) and perform device (not shown).
Second acquisition device obtains the feature relevant information of webpage to be assessed.
Specifically, the second acquisition device obtain the mode of feature relevant information of webpage to be assessed include but not limited to following any one:
1) multiple webpages and the feature relevant information thereof of self pre-stored are directly obtained;
2) the second acquisition device obtains from other equipment that it connects.
Preferably, the second acquisition device can obtain this webpage and feature relevant information thereof from one or more network equipments of publishing web page.
It is highly preferred that described webpage can be issued by this computer equipment self, and by other information relevant to this webpage via network monitoring or reception, determine the feature relevant information of this webpage.
Then, second determines that device is according to the feature relevant information obtained, it is determined that this target sample group corresponding to webpage to be assessed.
Wherein, described second determines that device is according to the feature relevant information obtained, determine the mode of this target sample group corresponding to webpage to be assessed and the implementation 2 of aforementioned first acquisition device 1) described in the computer equipment mode that multiple webpages is divided into multiple sample group according to the one or more feature relevant information of webpage same or similar, do not repeat them here.
Then, perform device, based on the webpage assessment models corresponding with this sample group of target, this webpage to be evaluated is performed evaluation operation.
The described assessment result that webpage to be assessed performs evaluation operation is used for the reception degree of the information indicating user to issue for webpage.Preferably, described assessment result includes following at least any one information:
1) probabilistic information that webpage is clicked;
2) the matching degree information of the user of webpage and predefined type.
Preferably, when described webpage to be assessed includes advertising message to be assessed, described Evaluated effect includes effect of advertising, the degree that described effect of advertising are accepted by user for indicating advertising message.It is highly preferred that described effect of advertising include but not limited to following at least any one:
1) probabilistic information that advertising message is clicked.
2) the matching degree information of the user of advertising message and predefined type.Wherein, described predefined type includes the classification after user is divided by the one or more user related information according to user.
Such as, the classification obtained after user being divided according to user's occupational information, wherein, described user's occupational information can be obtained by the data of family;Again such as, the classification etc. obtained after user being divided according to the age of user.
Wherein, described matching degree information for indicate advertising message and this predefined type user demand between agree with degree.Wherein, computer equipment can determine described matching degree information by following at least any of mode:
A) matching degree information is determined according to the description information of advertising message and the goodness of fit of user related information;Such as, the text message corresponding to audience determines matching degree information with the characters matching rate of self portrait partial content in subscriber data.
B) the matching degree information determined according to the conversion ratio of advertising message.Such as, directly using the conversion ratio of the advertising message matching degree information as this advertisement;Again such as, the conversion ratio relative to different types of user of advertising message is done the result after normalized as matching degree information etc..
Such as, the second acquisition device obtains three feature relevant informations of below advertisement Ad1 to be assessed:
Targeted customer's type of webpage belonging to advertising message: be positioned at the user of Area1, Area2 and Area3;
The clicked number of times of advertising message: 2324 times;
User's occupation of advertising message: company clerk.
Then second determines that device is according to above-mentioned three information, it is determined that the target cam group corresponding to advertisement Ad1 to be assessed is Ad_group1;
Then perform device and determine that the advertisement evaluations model corresponding to this target sample group Ad_group1 is Ad_Model1, wherein, the advertisement evaluations model Ad_Mode1 that this has been set up is for targeted customer's type information of webpage belonging to advertisement, the clicked number of times of advertising message, and click the user related information of the user of advertisement, the matching degree with the user being positioned at area Area1 of assessment advertising message, then perform device and adopt this initial assessment model Ad_Mode1, these three feature relevant informations execution evaluation operation based on the advertisement Ad1 to be evaluated that the second acquisition device obtains, obtain this advertisement Ad1 to be evaluated relative to the matching degree information of user being positioned at area Area1.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, any implementation that this webpage to be assessed performs evaluation operation based on the webpage assessment models corresponding with this sample group of target, should be included in the scope of the present invention.
According to the solution of the present invention, by performing Data Migration or model migration operation, based on current existing data or model, corresponding with target sample group, rational assessment models can be set up, to be estimated each advertising message belonging to target sample group determining its effect of advertising.Further, the advertisement evaluations device according to the present invention, it is possible to quickly realize convergence, it is established that stable assessment models, therefore can be applicable to the process of mass data preferably.Additionally, user can adjust the input situation of advertising message further according to effect of advertising, for instance, adjust the release time section of advertisement, to improve its clicking rate;Again such as, the targeted customer of advertisement is adjusted, to improve the matching degree etc. of advertisement and user, thus improve the efficiency of advertisement putting and the conversion ratio of advertisement.
Fig. 6 illustrates the structural representation of a kind of advertisement evaluations device for webpage is estimated in accordance with a preferred embodiment of the present invention.Advertisement evaluations device according to the present embodiment, described weight adjustment information includes each sample for reference group in described least one set sample for reference group and is respectively relative to the packet weight information of described target sample group.Advertisement evaluations device according to the present embodiment includes the first acquisition device 1, Weight Acquisition device 2, model acquisition device 3 and evaluation operation device 4.Wherein, described Weight Acquisition device 2 farther includes packet Weight Acquisition device 21, and described model acquisition device 3 farther includes the first adjusting apparatus 31 and the first son obtains 32.
First acquisition device 1 and evaluation operation device 4 are described in detail with reference to the embodiment shown in FIG. 5, and are incorporated herein with way of reference, repeat no more.
Preferably, described packet weight information includes each sample for reference group the first similarity relative to target sample group.
Preferably, when recanalization information of holding power includes that in described least one set sample for reference group, each sample for reference group is respectively relative to the packet weight information of described target sample group, packet Weight Acquisition device 21 each feature relevant information to each sample for reference group, obtain this stack features relevant information the first similarity relative to this feature relevant information of target sample group, to determine each reference group packet weight information relative to target sample group according to this first similarity.
Such as, each advertising message organized in sample for reference group and target sample group all comprises two feature relevant informations: user's occupational information and advertisement positional information in webpage.Further, the predetermined accounting weighted value of user's occupational information is 20%, and the predetermined accounting weight of the positional information in advertisement webpage belonging to it is 80%.Then packet Weight Acquisition device 21 first determines the often group sample for reference group value relative to the similarity Sim1 of target sample group according to user's occupational information, and further according to advertisement positional information in webpage, this determines the often group sample for reference group value relative to the similarity Sim2 of target sample group;And finally determine that often group sample for reference group is relative to first similarity (Sim1*20%+Sim2*80%) of target sample group, and the value obtained is made the packet weight information of this group sample for reference group.
The first adjusting apparatus 31 packet weight information according to each sample for reference group, is adjusted the feature relevant information of each webpage in each sample for reference group, to obtain the feature relevant information of each webpage after adjusting.
The first example according to the present invention, target sample group Group1 has m corresponding sample for reference group Group1_1, Group1_2 ..., Group1_m, and, this m sample for reference group is relative to the packet weight information respectively w of this target sample group1, W2, W3..., wm, wherein, the packet weight information of the sample for reference group of i-th group is Wi, wherein, i is natural number, and 1≤i≤m.The then first adjusting apparatus 31 packet weight according to each sample for reference group, the feature relevant information of each advertising message in this sample for reference group is adjusted, to obtain each group Group1_1 ', the Group1_2 of the feature relevant information after comprising adjustment ', ..., Group1_m '.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, any packet weight information according to each sample for reference group, the feature relevant information of each webpage in each sample for reference group is adjusted, to obtain the implementation of the feature relevant information of each webpage after adjusting, should be included in the scope of the present invention.
Then, the first sub-acquisition device 32 adopts predetermined training method to be trained based on the feature relevant information after the adjustment of described target sample group and described sample for reference group, with the webpage assessment models that the webpage obtained with described target sample group adapts.
Continue first example of the present invention is illustrated, first sub-acquisition device 32 is based on each group Group1, Group1_1 of obtaining ', Group1_2 ', ..., the feature relevant information of all advertising messages comprised in Group1_m ', predetermined support vector machine training method is adopted to be trained, to obtain the advertisement evaluations model corresponding with target sample group Group1.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, any feature relevant information based on after the adjustment of described target sample group and described sample for reference group adopts predetermined training method to be trained, with the implementation of the webpage assessment models that the webpage obtained with described target sample group adapts, should be included in the scope of the present invention.
Scheme according to the present embodiment, by using the Data Migration of reference group as target group data, achieve the extensive of target group data to a certain extent so that the model set up is when less relative to the deviation of former target group data, and the variation when it is extensive is also less.So that the webpage assessment models obtained has good effect when assessment and target sample group have other advertising messages of similar characteristic.Further, according to the scheme of the present embodiment owing to data need not be carried out iteration repeatedly, so processing mass data, for instance convergence can be rapidly achieved during the process of the advertising message of magnanimity, greatly improve the operational efficiency of computer equipment, there is greater advantage.
Fig. 7 illustrates the structural representation of a kind of advertisement evaluations device for webpage is estimated according to another preferred embodiment of the present invention.Weight adjustment information according to the present embodiment includes Model Weight information.Advertisement evaluations device according to the present embodiment includes the first acquisition device 1, Weight Acquisition device 2, model acquisition device 3 and evaluation operation device 4.Wherein, described Weight Acquisition device 2 farther includes initial acquisition device 22 and the first sub-Weight Acquisition 23, and described model acquisition device 3 farther includes the second sub-acquisition device 33.
First acquisition device 1 and evaluation operation device 4 are described in detail with reference to the embodiment shown in FIG. 5, and are incorporated herein with way of reference, repeat no more.
Initial acquisition device 22 obtains and described target sample group and each sample for reference group initial assessment model corresponding, that obtain based on predetermined training method respectively.
Specifically, described acquisition and described target sample group and each sample for reference group mode corresponding, initial assessment model that is that obtain based on predetermined training method respectively include but not limited to following any one:
1) initial acquisition device 22 obtains the pre-established initial assessment model not corresponding with each sample components.
Specifically, described initial acquisition device 22 obtain the mode of the pre-established initial assessment model not corresponding with each sample components include but not limited to following any one:
A) directly obtain initial acquisition device 22 self pre-stored of said computer equipment this is pre-
Each the initial assessment model not corresponding with each sample components set up.
B) obtained by other equipment being connected with initial acquisition device 22 said computer equipment
Take this pre-established each initial assessment model not corresponding with each sample components.
2) each sample group is adopted predetermined training method to be trained by initial acquisition device 22, to obtain the initial assessment model not corresponding with each sample components.
The second example according to the present invention, target sample group Group2 has the sample for reference group Group2_1 that p group is corresponding, Group2_2, ..., Group2_p, then this p group sample group is adopted preselected neural metwork training mode to be trained by initial acquisition device 22 respectively, to obtain initial assessment model M odel2_1 corresponding with each sample for reference group respectively, Model2_2 ..., Model2_p.Similarly, target sample group Group2 is adopted preselected nerve net training method to be trained by initial acquisition device 22, it is thus achieved that corresponding initial assessment model M odel2.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, any acquisition and described target sample group and each sample for reference group implementation corresponding, initial assessment model that is that obtain based on predetermined training method respectively, should be included in the scope of the present invention.
First sub-Weight Acquisition device 23 obtains the Model Weight information that the initial assessment model of each sample for reference group is respectively relative to the webpage of described target sample group.
Specifically, for each initial assessment model, first sub-Weight Acquisition device 23 adopts initial assessment model that the webpage in target sample group performs evaluation operation the accuracy according to assessment result, determines this initial assessment model Model Weight information relative to this target sample group.
Wherein, the accuracy of described assessment result can by obtaining after being compared by the actual information of assessment result Yu webpage.
Preferably, when described webpage includes advertising message, first sub-Weight Acquisition device 23 adopts initial assessment model that the advertising message in target sample group is performed evaluation operation to obtain the effect of advertising of each advertising message, and the accuracy according to the effect of advertising obtained, determine this initial assessment model Model Weight information relative to this target sample group.
Wherein, the accuracy of the assessment result of described advertising message can be obtained after being compared by the actual advertisement effect of effect of advertising assessment obtained and advertising message.
Such as, the effect of advertising of assessment gained of each advertising message in target sample group are compared by the first sub-Weight Acquisition device 23 by initial assessment model with the effect of advertising of its reality, have in evaluated obtained effect of advertising 80% with the difference of its actual effect of advertising in range of allowable error time, it is determined that the accuracy rate of this initial assessment model is 80%.
Preferably, the first sub-Weight Acquisition device 23 can by similar operation, it is thus achieved that the Model Weight information of target sample group self.
Preferably, the first sub-Weight Acquisition device 23 can directly using described accuracy as described Model Weight value.It is highly preferred that described Model Weight value can be pre-stored in this first sub-Weight Acquisition device 23 said computer equipment.It is highly preferred that described Model Weight value can prestore in other equipment being connected with this this first sub-Weight Acquisition device 23..
Continue the second example is illustrated, first sub-Weight Acquisition device 23 is respectively adopted p initial assessment model M odel2_1, Model2_2, ..., advertising message in target sample group Group2 is estimated by Model2_p, with obtain the multiple advertising messages in Group2 be based respectively on this each initial assessment model obtain effect of advertising, and by by assess gained effect of advertising compared with actual advertisement effect, to obtain the accuracy Ac of described p initial assessment model respectively1, Ac2..., Acp, then the first sub-Weight Acquisition device 23 is using this each accuracy as the Model Weight information of the initial assessment model of its correspondence.Similarly, first sub-Weight Acquisition device 23 adopts the initial assessment model M odel2 of target sample group Group2 that the advertising message in target sample group Group2 is estimated, it is based respectively on this initial assessment model M odel2 effect of advertising obtained obtaining the multiple advertising messages in target sample group Group2, and by the effect of advertising by assessing gained compared with actual advertisement effect, thus obtaining the Model Weight value Acg of this initial assessment model M odel2.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, the initial assessment model of each sample for reference group of any acquisition is respectively relative to the implementation of the Model Weight information of the webpage of described target sample group, should be included in the scope of the present invention.
The initial assessment model of described target group, according to the initial assessment model of each sample for reference group described and Model Weight information thereof, is adjusted by the second sub-acquisition device 33, to obtain the webpage assessment models adapted with described target sample group.
Preferably, the second sub-acquisition device 33 is according to the initial assessment model of each sample for reference group described and Model Weight information thereof, using the weighted mean of the initial assessment model of each sample for reference group described as described webpage assessment models.
Continuing the second example is illustrated, the second sub-acquisition device 33 determines the weighted mean that advertisement evaluations model is each initial assessment model corresponding for Group2, that is the final advertisement evaluations model M odel2 ' obtained including:
(Model2*Acg+Model2_1*Ac1+Model2_2*Ac2+...+Model2_p*Acp)/(Acg+Ac1+Ac2+...+Acp)。
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, the initial assessment model of each sample for reference group described in any basis and Model Weight information thereof, the initial assessment model of described target group is adjusted, to obtain the implementation of the webpage assessment models adapted with described target sample group, should be included in the scope of the present invention.
Preferred version according to the present embodiment, computer equipment obtains the initial assessment model corresponding with target sample group, and based on the initial assessment model that all webpages of described target sample group and each sample for reference group generate, and obtain the Model Weight information of these two initial assessment models respectively, to obtain the final webpage assessment models corresponding with target sample group.
Scheme according to the present embodiment, realizes less deviation by the initial assessment model corresponding with target sample group, and by the initial assessment model of other sample groups after adjusting in conjunction with weight, decreases the variation of model evaluation result.So that the final webpage assessment models obtained has the suitability widely and assessment result more accurately.Further, according to the migration operation of the present embodiment, each data need not being iterated operation, so convergence can be rapidly achieved, therefore, it is possible to be well applicable to the process of mass data, the assessment for the advertising message of current magnanimity can obtain good effect.
Fig. 8 illustrates the structural representation of a kind of advertisement evaluations device for webpage is estimated according to another preferred embodiment of the present invention.Weight adjustment information according to the present embodiment includes difference weight model.Advertisement evaluations device according to the present embodiment includes the first acquisition device 1, Weight Acquisition device 2, model acquisition device 3 and evaluation operation device 4.Wherein, described Weight Acquisition device 2 farther includes the second sub-Weight Acquisition device 24, and described model acquisition device 3 farther includes the 3rd sub-acquisition device 34 and the 4th sub-acquisition device 35.
First acquisition device 1 and evaluation operation device 4 are described in detail with reference to the embodiment shown in FIG. 5, and are incorporated herein with way of reference, repeat no more.
Second sub-Weight Acquisition device 24 is according to described target sample group, described least one set sample for reference group and described predetermined training method, it is thus achieved that for performing the difference weight model of migration operation.
Specifically, the second sub-Weight Acquisition device 24 is according to described target sample group, described least one set sample for reference group and described predetermined training method, it is thus achieved that the mode of difference weight model for performing migration operation include but not limited to following any one:
1) the second sub-Weight Acquisition device 24 obtains the pre-established difference weight model corresponding with each sample group and predetermined training method.
Specifically, the mode of the described pre-established difference weight model corresponding with each sample group and predetermined training method include but not limited to following any one:
A) second sub-Weight Acquisition device 24 self pre-stored of said computer equipment, this pre-established difference weight model corresponding with each sample group and predetermined training method are directly obtained.
B) by other equipment being connected with the computer equipment belonging to the second sub-Weight Acquisition device 24 obtain this pre-established difference weight model corresponding with each sample group and predetermined training method.
2) first training devices's (not shown) that the second sub-Weight Acquisition device 24 comprises first adopts described predetermined training method that the webpage of the least one set sample for reference group of described target sample group and correspondence thereof is trained, to obtain pre-evaluation model;Then, described target sample group is adopted described pre-evaluation model to continue executing with training by second training devices's (not shown) that the second sub-Weight Acquisition device 24 comprises, to obtain difference evaluation model;The difference that then the second sub-Weight Acquisition device 24 comprises determines that device (not shown) determines difference weight model according to described pre-evaluation model and described difference evaluation model.
nullThe 3rd example according to the present invention,First training devices adopts the predetermined neural metwork training mode q group sample for reference group Group3_1 to target sample group Group3 and correspondence thereof,Group3_2,...,Whole advertising messages that Group3_q comprises are trained,To obtain the pre-evaluation model W_Model based on all advertising messages,Then,Second training devices individually adopts the Group3 advertising message comprised that this pre-evaluation model W_Model proceeds training,To obtain difference evaluation model W_Model ',Then difference determines that device is according to the pre-evaluation model W_Model obtained and difference evaluation model W_Model ',Determine that the difference weight model corresponding with target sample group Group3 is | W_Model-W_Model ' |.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, any according to described target sample group, described least one set sample for reference group and described predetermined training method, obtain the implementation being used for performing the difference weight model of migration operation, should be included in the scope of the present invention.
3rd sub-acquisition device 34 adopts described predetermined training method that the webpage of described target sample group is trained, to obtain initial assessment model.
Continuing the 3rd example is illustrated, the 3rd sub-acquisition device 34 adopts predetermined neural metwork training mode that target sample group Group3 is trained, to obtain the initial assessment model M odel3 corresponding with target sample group Group3.
Described initial assessment model, according to described pre-evaluation model and described difference weight model, is adjusted by the 4th sub-acquisition device 35, with the webpage assessment models that the webpage obtained with this target sample group adapts.
Continue the 3rd example is illustrated, 4th sub-acquisition device 35 determines determined difference weight model | the W_Model-W_Model ' | of device 24 according to the second sub-weight, adjust the initial assessment model M odel3 corresponding with target sample group Group3 that the 3rd sub-acquisition device 34 obtains, to obtain final advertisement evaluations model M odel3 '=Model3+ | the W_Model-W_Model ' | after adjusting.
It should be noted that, the example above is only and technical scheme is better described, but not limitation of the present invention, those skilled in the art should understand that, the initial assessment model of each sample for reference group described in any basis and Model Weight information thereof, the initial assessment model of described target group is adjusted, to obtain the implementation of the webpage assessment models adapted with described target sample group, should be included in the scope of the present invention.
Scheme according to the present embodiment, the webpage assessment models obtained contains the initial assessment model corresponding with target sample group, current model is made to have less deviation, and, advertising message for current goal sample group, current initial assessment model is adjusted so that initial assessment model has better adaptability, thus being not susceptible to variation by difference weight model.Additionally, the scheme according to the present embodiment can be rapidly achieved convergence when processing mass data, alleviate the burden of computer equipment, decrease required computer equipment quantity.
The software program of the present invention can perform to realize steps described above or function by processor.Similarly, the software program of the present invention can be stored in computer readable recording medium storing program for performing (including the data structure being correlated with), for instance, RAM memory, magnetically or optically driver or floppy disc and similar devices.It addition, some steps of the present invention or function can employ hardware to realize, for instance, as coordinating with processor thus performing the circuit of each function or step.
It addition, the part of the present invention can be applied to computer program, for instance computer program instructions, when it is computer-executed, by the operation of this computer, it is possible to call or provide the method according to the invention and/or technical scheme.And call the programmed instruction of the method for the present invention, it is possibly stored in fixing or moveable record medium, and/or by broadcast or data stream in other signal bearing medias and be transmitted, and/or be stored in the working storage of the computer equipment run according to described programmed instruction.At this, include a device according to one embodiment of present invention, this device includes the memorizer for storing computer program instructions and for performing the processor of programmed instruction, wherein, when this computer program instructions is performed by this processor, trigger this plant running based on the method for aforementioned multiple embodiments according to the present invention and/or technical scheme.
It is obvious to a person skilled in the art that the invention is not restricted to the details of above-mentioned one exemplary embodiment, and when without departing substantially from the spirit of the present invention or basic feature, it is possible to realize the present invention in other specific forms.Therefore, no matter from which point, embodiment all should be regarded as exemplary, and be nonrestrictive, the scope of the invention rather than described above limits, it is intended that all changes in the implication of the equivalency dropping on claim and scope be included in the present invention.Any accompanying drawing labelling in claim should be considered as the claim that restriction is involved.Furthermore, it is to be understood that " including " word is not excluded for other unit or step, odd number is not excluded for plural number.Multiple unit or the device stated in system claims can also be realized by software or hardware by a unit or device.The first, the second word such as grade is used for representing title, and is not offered as any specific order.
Claims (21)
1. a computer implemented method for webpage is estimated, wherein, described webpage includes multinomial feature relevant information, said method comprising the steps of:
A obtains one group of target sample group of webpage assessment models to be set up and corresponding least one set sample for reference group, and wherein, the least one set sample for reference group of described target sample group and correspondence thereof includes multiple webpage respectively;
B is according to described target sample group and described least one set sample for reference group, it is thus achieved that in described least one set sample for reference group part or all refer to the sample group weight adjustment information relative to described target sample group;
The c described weight adjustment information corresponding to described target sample group, each sample for reference group and predetermined training method perform transfer learning operation, it is thus achieved that the webpage assessment models of described target sample group;
One or more webpages to be assessed are estimated by d based on described webpage assessment models.
2. method according to claim 1, wherein, described weight adjustment information includes each sample for reference group in described least one set sample for reference group and is respectively relative to the packet weight information of described target sample group, described packet weight information includes each sample for reference group comprising multiple advertising message weight information relative to target sample group, and step b includes:
-each feature relevant information to each sample for reference group, obtain the feature relevant information first similarity relative to the feature relevant information of target sample group of this reference cam group, to determine each reference group packet weight information relative to target sample group according to this first similarity;
Wherein, described step c includes:
-packet weight information according to each sample for reference group, is adjusted the feature relevant information of each webpage in each sample for reference group, to obtain the feature relevant information of each webpage after adjusting;
-adopt predetermined training method to be trained based on the feature relevant information after the adjustment of described target sample group and described sample for reference group, with the webpage assessment models that the webpage obtained with described target sample group adapts.
3. method according to claim 1, wherein, described weight adjustment information includes Model Weight information, described Model Weight information includes reference group model that each sample for reference group set up by the predetermined training method weight information relative to target sample group, and described step b comprises the following steps:
-obtain and described target sample group and each sample for reference group initial assessment model corresponding, that obtain based on predetermined training method respectively;
-obtain the Model Weight information that the initial assessment model of each sample for reference group is respectively relative to the webpage of described target sample group;
Wherein, step c comprises the following steps:
-according to the initial assessment model of each sample for reference group described and Model Weight information thereof, the initial assessment model of described target group is adjusted, to obtain the webpage assessment models adapted with described target sample group.
4. method according to claim 1, wherein, described weight adjustment information includes difference weight model, and described difference weight model includes the model for determining the weight difference that target group model is adjusted, and described step b comprises the following steps:
B1 is according to described target sample group, described least one set sample for reference group and described predetermined training method, it is thus achieved that for performing the difference weight model of migration operation;
Wherein, described step c comprises the following steps:
-adopt described predetermined training method that the webpage of described target sample group is trained, to obtain initial assessment model;
-according to pre-evaluation model and described difference weight model, described initial assessment model is adjusted, with the webpage assessment models that the webpage obtained with this target sample group adapts.
5. method according to claim 4, wherein, described step b1 comprises the following steps:
-adopt described predetermined training method to be trained the least one set sample for reference group of described target sample group and correspondence thereof, to obtain pre-evaluation model;
-based on described pre-evaluation model, continue executing with training based on described target sample group, to obtain difference evaluation model;
-determine difference weight model according to described pre-evaluation model and described difference evaluation model.
6. method according to any one of claim 1 to 5, wherein, described step a comprises the following steps:
-according to one or more feature relevant information, multiple webpages are divided into multiple sample group;
-by the many web pages divided selecting least one set as described target sample group;
Wherein, to selected each target sample group, following steps are also performed:
A1 is by selecting one or more groups as the sample for reference group corresponding with this target sample group in other web pages remaining.
7. method according to claim 6, wherein, described step a1 comprises the following steps:
-determine the second similarity between one or more groups and this target sample group in other web pages remaining;
-according to described second similarity, and select rule based on predetermined, by one or more groups webpage described selects the least one set webpage similar to described target sample group, as the sample for reference group of described target sample group.
8. method according to any one of claim 1 to 5, wherein, described step d comprises the following steps:
-obtain the feature relevant information of webpage to be assessed;
-according to the feature relevant information obtained, it is determined that this target sample group corresponding to webpage to be assessed;
-based on the webpage assessment models corresponding with this sample group of target, this webpage to be assessed is performed evaluation operation.
9. the method according to described claim 8, wherein, the described assessment result that webpage to be assessed performs evaluation operation includes following at least any one information:
The probabilistic information that-webpage is clicked;
The matching degree information of the user of-webpage and predefined type.
10. method according to any one of claim 1 to 5, wherein, described feature relevant information includes following at least any one information:
The object-related information promoting object that-webpage comprises;
The clicked information of-webpage;
The user related information of the user of-webpage clicking;
-webpage represent type information.
11. for the webpage apparatus for evaluating that webpage is estimated, wherein, described webpage includes multinomial feature relevant information, and described webpage apparatus for evaluating includes:
First acquisition device, for obtaining one group of target sample group of webpage assessment models to be set up and corresponding least one set sample for reference group, wherein, the least one set sample for reference group of described target sample group and correspondence thereof includes multiple webpage respectively;
Weight Acquisition device, for according to described target sample group and described least one set sample for reference group, it is thus achieved that part or all refer to the sample group weight adjustment information relative to described target sample group in described least one set sample for reference group;
Model acquisition device, performs transfer learning operation for the described weight adjustment information corresponding to described target sample group, each sample for reference group and predetermined training method, it is thus achieved that the webpage assessment models of described target sample group;
Evaluation operation device, for being estimated one or more webpages to be assessed based on described webpage assessment models.
12. webpage apparatus for evaluating according to claim 11, wherein, described weight adjustment information includes each sample for reference group in described least one set sample for reference group and is respectively relative to the packet weight information of described target sample group, described packet weight information includes each sample for reference group comprising multiple advertising message weight information relative to target sample group, wherein, described Weight Acquisition device includes:
Packet Weight Acquisition device, for each the feature relevant information to each sample for reference group, obtain the feature relevant information first similarity relative to the feature relevant information of target sample group of this reference cam group, to determine each reference group packet weight information relative to target sample group according to this first similarity;
Wherein, described model acquisition device includes:
First adjusting apparatus, for the packet weight information according to each sample for reference group, is adjusted the feature relevant information of each webpage in each sample for reference group, to obtain the feature relevant information of each webpage after adjusting;
First sub-acquisition device, for adopting predetermined training method to be trained based on the feature relevant information after the adjustment of described target sample group and described sample for reference group, with the webpage assessment models that the webpage obtained with described target sample group adapts.
13. webpage apparatus for evaluating according to claim 11, wherein, described weight adjustment information includes Model Weight information, described Model Weight information includes reference group model that each sample for reference group set up by the predetermined training method weight information relative to target sample group, and described Weight Acquisition device includes:
Initial acquisition device, for obtaining and described target sample group and each sample for reference group initial assessment model corresponding, that obtain based on predetermined training method respectively;
First sub-Weight Acquisition device, the initial assessment model for obtaining each sample for reference group is respectively relative to the Model Weight information of the webpage of described target sample group;
Wherein, described model acquisition device includes:
Second sub-acquisition device, for the initial assessment model according to each sample for reference group described and Model Weight information thereof, is adjusted the initial assessment model of described target group, to obtain the webpage assessment models adapted with described target sample group.
14. webpage apparatus for evaluating according to claim 11, wherein, described weight adjustment information includes difference weight model, and described difference weight model includes the model for determining the weight difference that target group model is adjusted, and described Weight Acquisition device includes:
Second sub-Weight Acquisition device, for according to described target sample group, described least one set sample for reference group and described predetermined training method, it is thus achieved that for performing the difference weight model of migration operation;
Wherein, described model acquisition device includes:
3rd sub-acquisition device, for adopting described predetermined training method that the webpage of described target sample group is trained, to obtain initial assessment model;
4th sub-acquisition device, for according to pre-evaluation model and described difference weight model, being adjusted described initial assessment model, with the webpage assessment models that the webpage obtained with this target sample group adapts.
15. webpage apparatus for evaluating according to claim 14, wherein, described second sub-Weight Acquisition device includes:
First training devices, for adopting described predetermined training method to be trained the least one set sample for reference group of described target sample group and correspondence thereof, to obtain pre-evaluation model;
Second training devices, for based on described pre-evaluation model, continuing executing with training based on described target sample group, to obtain difference evaluation model;
Difference determines device, for determining difference weight model according to described pre-evaluation model and described difference evaluation model.
16. the webpage apparatus for evaluating according to any one of claim 11 to 15, wherein, described first acquisition device includes:
Divide device, for multiple webpages being divided into multiple sample group according to one or more feature relevant information;
First selects device, for by selecting least one set as described target sample group in the many web pages divided;
Wherein, to selected each target sample group, described first acquisition device also includes:
Second selects device, for by selecting one or more groups as the sample for reference group corresponding with this target sample group in other web pages remaining.
17. webpage apparatus for evaluating according to claim 16, wherein, described second selects device to include:
First determines device, for determining the second similarity between one or more groups and this target sample group in other web pages remaining;
Son selects device, is used for according to described second similarity, and selects rule based on predetermined, by selecting the least one set webpage similar to described target sample group in one or more groups webpage described, as the sample for reference group of described target sample group.
18. the webpage apparatus for evaluating according to any one of claim 11 to 15, wherein, described evaluation operation device includes:
Second acquisition device, for obtaining the feature relevant information of webpage to be assessed;
Second determines device, for according to the feature relevant information obtained, it is determined that this target sample group corresponding to webpage to be assessed;
Perform device, for this webpage to be assessed being performed evaluation operation based on the webpage assessment models corresponding with this sample group of target.
19. the webpage apparatus for evaluating according to described claim 18, wherein, the described assessment result that webpage to be assessed performs evaluation operation includes following at least any one information:
The probabilistic information that-webpage is clicked;
The matching degree information of the user of-webpage and predefined type.
20. the webpage apparatus for evaluating according to any one of claim 11 to 15, wherein, described feature relevant information includes following at least any one information:
The object-related information promoting object that-webpage comprises;
The clicked information of-webpage;
The user related information of the user of-webpage clicking;
-webpage represent type information.
21. a computer equipment, wherein, this computer equipment includes such as webpage apparatus for evaluating as described at least one in claim 11 to 20.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210343159.4A CN102880688B (en) | 2012-09-14 | 2012-09-14 | A kind of method for webpage is estimated, device and equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210343159.4A CN102880688B (en) | 2012-09-14 | 2012-09-14 | A kind of method for webpage is estimated, device and equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102880688A CN102880688A (en) | 2013-01-16 |
CN102880688B true CN102880688B (en) | 2016-07-27 |
Family
ID=47482014
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210343159.4A Active CN102880688B (en) | 2012-09-14 | 2012-09-14 | A kind of method for webpage is estimated, device and equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102880688B (en) |
Families Citing this family (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104751234B (en) * | 2013-12-31 | 2018-10-19 | 华为技术有限公司 | A kind of prediction technique and device of user's assets |
WO2017069735A1 (en) | 2015-10-20 | 2017-04-27 | Viasat, Inc. | Hint model updating using automated browsing clusters |
CN105450766A (en) * | 2015-12-07 | 2016-03-30 | 浪潮集团有限公司 | FPGA rapid time sequence convergence system |
CN105678583B (en) * | 2016-01-07 | 2021-05-07 | 腾讯科技(深圳)有限公司 | Information distribution method and information distribution device |
CN106227743B (en) * | 2016-07-12 | 2019-09-24 | 精硕科技(北京)股份有限公司 | Advertisement target group touching reaches ratio estimation method and device |
CN107437111B (en) * | 2017-07-31 | 2020-07-14 | 杭州朗和科技有限公司 | Data processing method, medium, device and computing equipment based on neural network |
CN108960274A (en) * | 2018-05-07 | 2018-12-07 | 浙江大学 | A kind of Active Learning Method for the accessible inspection assessment of webpage information |
CN108898432B (en) * | 2018-06-25 | 2022-05-13 | 武汉斗鱼网络科技有限公司 | Advertisement putting effect evaluation method and device and electronic equipment |
CN109064043B (en) * | 2018-08-14 | 2020-12-08 | 北京猿力未来科技有限公司 | Evaluation method, evaluation device, computing equipment and storage medium |
CN109145978A (en) * | 2018-08-15 | 2019-01-04 | 大连海事大学 | A kind of weak relevant cluster method of the feature of shoe sole print image |
CN109726762B (en) * | 2018-12-29 | 2021-08-27 | 北京达佳互联信息技术有限公司 | Video type determination method and device, electronic equipment and storage medium |
CN110046952B (en) * | 2019-01-30 | 2021-12-10 | 创新先进技术有限公司 | Recommendation model training method and device, and recommendation method and device |
CN109784761A (en) * | 2019-01-31 | 2019-05-21 | 中国互联网络信息中心 | Domain name ranking method, device, electronic equipment and storage medium based on block chain |
CN111985751B (en) * | 2019-05-23 | 2023-09-26 | 百度在线网络技术(北京)有限公司 | Human-computer chat experience assessment system |
CN110472885A (en) * | 2019-08-22 | 2019-11-19 | 华南师范大学 | A kind of website assessment system and its working method |
CN115086705A (en) * | 2021-03-12 | 2022-09-20 | 北京字跳网络技术有限公司 | Resource preloading method, device, equipment and storage medium |
CN113190779B (en) * | 2021-05-08 | 2023-07-28 | 北京百度网讯科技有限公司 | Webpage evaluation method and device |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101382939A (en) * | 2008-10-23 | 2009-03-11 | 浙江大学 | Web page text individuation search method based on eyeball tracking |
CN102346899A (en) * | 2011-10-08 | 2012-02-08 | 亿赞普(北京)科技有限公司 | Method and device for predicting advertisement click rate based on user behaviors |
CN102508859A (en) * | 2011-09-29 | 2012-06-20 | 北京亿赞普网络技术有限公司 | Advertisement classification method and device based on webpage characteristic |
CN102663617A (en) * | 2012-03-20 | 2012-09-12 | 亿赞普(北京)科技有限公司 | Method and system for prediction of advertisement clicking rate |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8756184B2 (en) * | 2009-12-01 | 2014-06-17 | Hulu, LLC | Predicting users' attributes based on users' behaviors |
-
2012
- 2012-09-14 CN CN201210343159.4A patent/CN102880688B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101382939A (en) * | 2008-10-23 | 2009-03-11 | 浙江大学 | Web page text individuation search method based on eyeball tracking |
CN102508859A (en) * | 2011-09-29 | 2012-06-20 | 北京亿赞普网络技术有限公司 | Advertisement classification method and device based on webpage characteristic |
CN102346899A (en) * | 2011-10-08 | 2012-02-08 | 亿赞普(北京)科技有限公司 | Method and device for predicting advertisement click rate based on user behaviors |
CN102663617A (en) * | 2012-03-20 | 2012-09-12 | 亿赞普(北京)科技有限公司 | Method and system for prediction of advertisement clicking rate |
Also Published As
Publication number | Publication date |
---|---|
CN102880688A (en) | 2013-01-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102880688B (en) | A kind of method for webpage is estimated, device and equipment | |
US20190147461A1 (en) | Methods and apparatus to estimate total audience population distributions | |
CN104035926B (en) | A kind of dispensing of internet information and system | |
US11593860B2 (en) | Method, medium, and system for utilizing item-level importance sampling models for digital content selection policies | |
US10262336B2 (en) | Non-converting publisher attribution weighting and analytics server and method | |
CN106462560A (en) | Systems and methods for optimizing content layout using behavior metrics | |
CN103248705B (en) | Server, client and method for processing video frequency | |
CN110796477A (en) | Advertisement display method and device, electronic equipment and readable storage medium | |
CN105023165A (en) | Method, device and system for controlling release tasks in social networking platform | |
CN104657372A (en) | Page operation data processing method and device | |
CN104331459A (en) | Online learning-based network resource recommendation method and device | |
CN103049452A (en) | Method and device for performing application sequencing based on estimated download rate | |
CN105590240A (en) | Discrete calculating method of brand advertisement effect optimization | |
CN108777701A (en) | A kind of method and device of determining receiver | |
US10853428B2 (en) | Computing a ranked feature list for content distribution in a first categorization stage and second ranking stage via machine learning | |
CN113961830A (en) | System and method for segmenting client sessions of a website using web scripts | |
JP2011227721A (en) | Interest extraction device, interest extraction method, and interest extraction program | |
CN105574198A (en) | Column recommendation method and device | |
US9922340B1 (en) | Evaluating attribution models based on simulated activity streams | |
CN103744938A (en) | Method and device for generating user page corresponding to target system | |
CN108694174B (en) | Content delivery data analysis method and device | |
CN104735117A (en) | Information processing method, device and system | |
CN111260416B (en) | Method and device for determining associated user of object | |
CN103793427A (en) | Method and equipment for determining optimized presenting keyword of object presenting keyword | |
CN109697203B (en) | Index transaction analysis method and device, computer storage medium, and computer device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |