CN109684554A - The determination method and news push method of the potential user of news - Google Patents
The determination method and news push method of the potential user of news Download PDFInfo
- Publication number
- CN109684554A CN109684554A CN201811601767.4A CN201811601767A CN109684554A CN 109684554 A CN109684554 A CN 109684554A CN 201811601767 A CN201811601767 A CN 201811601767A CN 109684554 A CN109684554 A CN 109684554A
- Authority
- CN
- China
- Prior art keywords
- news
- user
- pushed
- portrait
- target user
- 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.)
- Granted
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application involves the determination method of the potential user of news push method and news a kind of, news push method includes: the user's portrait for obtaining each target user;Each user is drawn a portrait and inputs corresponding with respectively news to be pushed each click prediction model, news to be pushed is corresponding, and to click prediction model to have clicked the user for the sample of users that the user of the sample of users of news push draws a portrait for positive sample, to have received but not click on the news to be pushed and draw a portrait be that negative sample progress model training obtains;By each click prediction model, the probability for predicting that each target user clicks on respectively news to be pushed of being drawn a portrait according to the user of each target user;According to each probability, the potential user of news respectively to be pushed is determined from each target user;It will respectively news to be pushed be associated with respective potential user;News is pushed to each target user according to the associated news to be pushed of each target user.The dependence for the accuracy that the accuracy that the scheme of the application can reduce news push draws a portrait to news.
Description
Technical field
This application involves field of computer technology, more particularly to a kind of potential user of news determination method, apparatus,
News push method, apparatus, computer readable storage medium and computer equipment.
Background technique
It is needed in the epoch of information explosion in order to help user's reduction to get the time cost of oneself interested news
The preferred news of user is excavated from magnanimity news, carry out news push accordingly.
In traditional approach, for each news to be pushed, the news to be pushed manually is determined by news editor personnel
News portrait, then news portrait is matched one by one with the user of each target user portrait, its user portrait is new with this
Potential user of the target user for drawing a portrait and matching as the news to be pushed is heard, carries out news push accordingly.However, once new
It hears portrait and deviation occurs, identified potential user is just inaccurate, this, which will lead to, can not accurately carry out news push.As it can be seen that
The accuracy of news push has serious dependence to the accuracy that news is drawn a portrait under traditional approach.
Summary of the invention
Based on this, it is necessary to have seriously for the accuracy of news push in traditional approach to the accuracy that news is drawn a portrait
The technical issues of dependence, provides the determination method, apparatus of the potential user of news a kind of, news push method, apparatus, calculates
Machine readable storage medium storing program for executing and computer equipment.
A kind of news push method, comprising:
Obtain user's portrait of each target user;
Each user is drawn a portrait, each click prediction model corresponding with respectively news to be pushed is inputted;Wherein, described
The corresponding prediction model of clicking of news to be pushed is the sample that is positive with the user's portrait for having clicked the sample of users of the news to be pushed
Originally, user's portrait of the sample of users to have received but not click on the news to be pushed is that negative sample progress model training obtains;
It by each click prediction model, is drawn a portrait according to the user of each target user, predicts that each target is used
Family clicks on the probability of respectively news to be pushed;
According to each probability, the potential user of each news to be pushed is determined from each target user;
Each news to be pushed is associated with respective potential user;
According to each associated news to be pushed of target user, news is pushed to each target user.
A kind of determination method of the potential user of news, comprising:
Obtain user's portrait of each target user;
Each user is drawn a portrait, click prediction model corresponding with news to be pushed is inputted;The click prediction model
Be take the user's portrait for having clicked the sample of users of the news to be pushed as positive sample, it is described wait push away to have received but not click on
Sending the user of the sample of users of news to draw a portrait is that negative sample progress model training obtains;
It by the click prediction model, is drawn a portrait according to the user of each target user, predicts each target user
Click on the probability of the news to be pushed;
According to each probability, the potential user of the news to be pushed is determined from each target user.
A kind of news push device, comprising:
First user, which draws a portrait, obtains module, and the user for obtaining each target user draws a portrait;
First portrait input module inputs corresponding with respectively news to be pushed each for each user to draw a portrait
Click prediction model;Wherein, the corresponding prediction model of clicking of the news to be pushed is the sample to have clicked the news to be pushed
User's portrait of sample of users of the user's portrait of this user for positive sample, to have received but not click on the news to be pushed is negative
Sample carries out model training and obtains;
First probability determination module is used for by each click prediction model, according to the user of each target user
Portrait predicts that each target user clicks on the probability of respectively news to be pushed;
First potential user's determining module, it is each described for being determined from each target user according to each probability
The potential user of news to be pushed;
First relating module, for each news to be pushed to be associated with respective potential user;
News push module, for according to each associated news to be pushed of target user, to each target user
Push news.
A kind of determining device of the potential user of news, comprising:
Second user portrait obtains module, and the user for obtaining each target user draws a portrait;
Second portrait input module inputs click prediction corresponding with news to be pushed for each user to draw a portrait
Model;It is described click prediction model and be with the user's portrait for having clicked the sample of users of the news to be pushed positive sample, with
The user's portrait for having received but not clicked on the sample of users of the news to be pushed is that negative sample progress model training obtains;
Second probability determination module, for being drawn according to the user of each target user by the click prediction model
Picture predicts that each target user clicks on the probability of the news to be pushed;
Second potential user's determining module, for according to each probability, determined from each target user it is described to
Push the potential user of news.
A kind of computer equipment, including memory and processor, the memory are stored with computer program, the calculating
When machine program is executed by the processor, so that the step of processor executes the above method.
According to scheme as described above, the user of each target user is drawn a portrait, is inputted to have clicked the sample of news to be pushed
User's portrait of this user be positive sample, sample of users to have received but not click on news to be pushed user's portrait be negative sample
The click prediction model that this progress model training obtains, then by clicking prediction model, drawn a portrait according to the user of each target user
It predicts that each target user clicks on the probability of news to be pushed, and then according to each probability, determines from each target user wait push away
Send the potential user of news.In this way, can determine that the potential of news to be pushed without the news portrait for relying on news to be pushed
User significantly reduces the dependence for the accuracy that the accuracy of news push draws a portrait to news.
Detailed description of the invention
Fig. 1 is the applied environment figure of news push method in one embodiment;
Fig. 2 is the applied environment figure of news push method in one embodiment;
Fig. 3 is the flow diagram of news push method in one embodiment;
Fig. 4 is that user draws a portrait in one embodiment to input the schematic diagram for clicking prediction model;
Fig. 5 is the schematic diagram that the output result of prediction model is clicked in one embodiment;
Fig. 6 is will news be pushed and the associated schematic diagram of potential user in one embodiment;
Fig. 7 is in one embodiment by the schematic diagram of the associated news to be pushed of target user;
Fig. 8 is the interface schematic diagram of news client in one embodiment;
Fig. 9 is the interface schematic diagram for sending news to be pushed in one embodiment user's notification bar;
Figure 10 is the schematic illustration that user's portrait is updated in one embodiment;
Figure 11 is the schematic diagram of XGBoost model inner workings in one embodiment;
Figure 12 is the timing diagram of news push method in one embodiment;
Figure 13 is the flow diagram of the determination method of the potential user of news in one embodiment;
Figure 14 is the structural block diagram of news push device in one embodiment;
Figure 15 is the structural block diagram of the determining device of the potential user of news in one embodiment;
Figure 16 is the structural block diagram of computer equipment in one embodiment.
Specific embodiment
It is with reference to the accompanying drawings and embodiments, right in order to which the objects, technical solutions and advantages of the application are more clearly understood
The application is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the application, and
It is not used in restriction the application.
It should be noted that term " first " used in this application, " second " etc. are for making to similar object
Differentiation in name, but these objects itself should not be limited by these terms.
Term "and/or" used in this application includes any and all of one or more relevant listed items
Combination.
The news push method that each embodiment of the application provides, can be applied in application environment as shown in Figure 1.The application
Environment can be related to user terminal 110 and news push-delivery system 120.User terminal 110 and news push-delivery system 120 can pass through network
Connection.
Specifically, news push-delivery system 120 obtains user's portrait of each target user;Then, each user is drawn a portrait and is inputted
Each click prediction model corresponding with respectively news to be pushed, wherein the corresponding click prediction model of news to be pushed is with point
The user's portrait for hitting the sample of users of the news to be pushed is positive sample, the sample to have received but not click on the news to be pushed
User's portrait of user is that negative sample progress model training obtains;Again by each click prediction model, according to each target user's
User's portrait, predicts that each target user clicks on the probability of respectively news to be pushed;In turn, it according to each probability, is used from each target
The respectively potential user of news to be pushed respectively is determined in family, and will respectively news to be pushed be associated with respective potential user;Then,
According to the associated news to be pushed of each target user, news is pushed to user terminal 110 corresponding to each target user.
Wherein, the server cluster that news push-delivery system 120 can be formed with multiple servers is realized.For example, such as Fig. 2 institute
Show, news push-delivery system 120 may include that (such as HDFS cluster, i.e. Hadoop are distributed for distributed storage server cluster 1202
File system, can be used for generating and storing each target user user portrait), potential user determine that server 1204 (can be used for
Determine the potential user of news to be pushed), news evaluating server 1206 (can be used for determining respectively each target user it is associated to
Push the recommendation index of news) and news the distribution server 1208 (for pushing corresponding recommendation to each target user
The server of news).
Accordingly, user terminal 110 corresponding with each target user can be by historical user's row of corresponding target user
Distributed storage server cluster 1202 is sent to for data.In turn, distributed storage server cluster 1202 is according to each target
Historical user's behavioral data of user generates respective user and draws a portrait and store.Then, potential user determine server 1204 from
Distributed storage server cluster 1202 obtain each target user user portrait, then execute by each user draw a portrait input with respectively to
Push the corresponding each click prediction model of news to will respectively news to be pushed is associated with respective potential user the step of, and
Association results are sent to news evaluating server 1206.In turn, news evaluating server 1206 determines each according to association results
The news to be pushed that target user is respectively associated, then be directed to each target user, determine respectively the target user it is associated to
The recommendation index of news is pushed, and index will be recommended to be sent to news the distribution server 1208.Then, news the distribution server
1208 are directed to each target user, according to the recommendation index of the associated news respectively to be pushed of the target user, use from the target
Family is associated respectively wait push in news, filters out the corresponding recommendation news of the target user, and the recommendation news push is given should
Target user.Wherein, the communication between user terminal 110 and distributed storage server cluster 1202 and news distribution clothes
The communication being engaged between device 1208 and user terminal 110 can be realized based on network communication apparatus 130.
Alternatively, news push-delivery system 120 can also be realized with independent server.The independent server executes alone from obtaining
Take the user of each target user draw a portrait to by corresponding the recommendations news push of each target user to corresponding user terminal 110
Series of steps.
In addition, user terminal 110 specifically may include mobile phone, tablet computer, laptop, desktop computer, individual digital
At least one of assistant, wearable device etc., but not limited to this.
In one embodiment, as shown in figure 3, providing a kind of news push method.In this way by above-mentioned Fig. 1
News push-delivery system 120 is illustrated for executing.This method may include following steps S302 to S312.
S302 obtains user's portrait of each target user.
Target user is the user for needing to push news to it.Target user can determine according to actual needs.For example, mesh
Mark user may include all registered users of news application program, and the backstage record of news application program has the use of user identifier
Family is registered user, and user identifier can be used for unique identification user, such as User ID (Identification, identity),
By taking " Tencent's news " application program as an example, it is assumed that its a total of N1 registered users can then make this N1 registered users
For target user.For another example, target user, which may also comprise in all registered users of application program, meets user's screening conditions
Certain customers, user's screening conditions can set according to actual needs, for example user's screening conditions may include User Status to live
Jump state, still by taking " Tencent's news " application program as an example, it is assumed that there are N2 any active ues in N1 registered users, then can sieve
This N2 any active ues is selected as target user.
User's portrait, is the user preference description information taken out according to historical user's behavioral data of target user,
It can be used for characterizing target user to the preference of each predetermined news category and each predetermined news label.Predetermined news category and pre-
Determining news label can preset according to actual needs.For example, predetermined news category may include " amusement " classification, " society " class
Not, " sport " classification and " science and technology " classification etc., predetermined news label may include content be the news label of " people's police ", content is
The news label and content of " Fan Bingbing " are the news label etc. of " millet ".
S304 draws a portrait each user, inputs each click prediction model corresponding with respectively news to be pushed.
News to be pushed is the news for needing to be pushed to target user.News to be pushed can be with prediction model is clicked
One-to-one corresponding relationship.The corresponding click prediction model of news to be pushed, being should be wait push for predicting that target user clicks
The algorithm model of the probability of news can be drawn a portrait with the user of target user to input, and clicking to target user should be wait push
The probability of news is predicted and is exported.
In the present embodiment, the user of each target user is drawn a portrait, inputs each point corresponding with respectively news to be pushed
Hit prediction model.For example, as shown in Figure 4 (for ease of description, hereinafter target user and news to be pushed are carried out with peanut
Citing, but the number of target user and news to be pushed can be any positive integer equal to or more than 1 in practical application), respectively
Target user is respectively target user U1, U2, U3 and U4, and respectively news to be pushed is respectively News1, News2 and News3, then will
The user portrait UP3 and target of user's portrait UP1 of target user U1, user's portrait UP2 of target user U2, target user U3
The user portrait UP4 of user U4, inputs the corresponding click prediction model CPM1 of news News1 to be pushed;By user draw a portrait UP1,
UP2, UP3 and UP4 input the corresponding click prediction model CPM2 of news News2 to be pushed;Also, by user draw a portrait UP1,
UP2, UP3 and UP4 input the corresponding click prediction model CPM3 of news News3 to be pushed.
Specifically, for each news to be pushed, the news mark of the news to be pushed can be first obtained, and new according to this
Hearing mark poll whether there is click prediction model corresponding with the news to be pushed.If it exists, then it loads and is somebody's turn to do wait push newly
Corresponding click prediction model is heard, and each user is drawn a portrait and inputs the click prediction model;If it does not exist, then without pushing,
But continue waiting for, additionally " there is no click prediction model " can be recorded as not push reason.
It should be noted that before loading corresponding with news to be pushed click prediction model, can first train obtain to
The corresponding click prediction model of push news simultaneously stores.It is subsequent when needing using the click prediction model, it can directly read and deposit
Click prediction model of storage, without temporarily carrying out model training again.
The corresponding click prediction model of news to be pushed can be to have clicked the use of the sample of users of the news to be pushed
Family portrait be positive sample and with received but do not clicked on (i.e. only expose) should the user of sample of users of news to be pushed draw a portrait and be
Negative sample carries out model training and obtains.Specifically, the news to be pushed of training click prediction model, news are needed for each
The news to be pushed first can be handed down to several sample of users by supplying system, then receive the use that each sample of users returns respectively
(the sample of users behavioral data of sample of users can be used for characterizing whether the sample of users clicks this wait push newly to family behavioral data
Hear), it regard user's portrait that its user behavior data characterization has clicked the sample of users of the news to be pushed as positive sample, and will
Its user behavior data characterization only receive news push but and do not click on the news to be pushed sample of users user picture
As being used as negative sample, and then model training is carried out according to each positive sample and each negative sample and obtains the corresponding click of the news to be pushed
Prediction model.
It should be noted that for the sample (user of i.e. any sample of users draws a portrait) that any training pattern uses, it should
Sample is corresponding with a standard results, which can be used for characterizing the probability that the sample of users clicks the news to be pushed.
For positive sample, standard results can be set as 1, can be used to characterization respective sample user and click the probability of the news to be pushed can be
1;For negative sample, standard results can be set as 0, can be used to characterization respective sample user click the probability of the news to be pushed can
It is 0.
For example, the data mode for the sample that training pattern uses can be (xi, yi), xiIndicate i-th of sample of users
User's portrait, yiIndicate the corresponding standard results of the sample, can be used to i-th of sample of users click of characterization should news be pushed
Probability.If xiFor positive sample, then yiIt can be 1;If xiFor negative sample, then yiIt can be 0.
S306 draws a portrait according to the user of each target user by each click prediction model, predicts each target user point respectively
Hit the probability of respectively news to be pushed.
It can be directed to each news to be pushed, by the corresponding click prediction model of the news to be pushed, according to each target
The user of user draws a portrait, and predicts that each target user clicks on the probability of the news to be pushed.
Aforementioned exemplary is accepted, as shown in figure 5, drawing a portrait by clicking prediction model CPM1 according to the user of target user U1
UP1, prediction target user U1 click the probability of news News1 to be pushed;By clicking prediction model CPM1, according to target user
User the portrait UP2, prediction target user U2 of U2 clicks the probability of news News1 to be pushed;By clicking prediction model CPM1,
The probability of news News1 to be pushed is clicked according to the user of target user U3 portrait UP3, prediction target user U3;Also, pass through
Prediction model CPM1 is clicked, news to be pushed is clicked according to the user of target user U4 portrait UP4, prediction target user U4
The probability of News1.
Similarly, by clicking prediction model CPM2, target user U1 is predicted according to the user of target user U1 portrait UP1
Click the probability of news News2 to be pushed, the UP2 prediction target user U2 that draws a portrait according to the user of target user U2 is clicked wait push
The probability of news News2, the UP3 prediction target user U3 that drawn a portrait according to the user of target user U3 click news News2 to be pushed
Probability and according to the user of target user U4 draw a portrait UP4 prediction target user U4 click the general of news News2 to be pushed
Rate.
Also, by clicking prediction model CPM3, target user U1 point is predicted according to the user of target user U1 portrait UP1
Hit the probability of news News3 to be pushed, the UP2 prediction target user U2 that draws a portrait according to the user of target user U2 is clicked wait push newly
Hear the probability of News3, the UP3 prediction target user U3 that draws a portrait according to the user of target user U3 clicks news News3's to be pushed
Probability and the probability that news News3 to be pushed is clicked according to the user of target user U4 portrait UP4 prediction target user U4.
S308 determines the potential user of news respectively to be pushed according to each probability from each target user.
Potential user, be clicked in each target user news to be pushed probability meet probability screening conditions target use
Family.Probability screening conditions can be set according to actual needs.Usually, it is intended to conditional filtering be screened by probability and go out to click wait push newly
The biggish target user of the probability of news, the potential user as the news to be pushed.
For example, probability screening conditions may include being equal to or more than probability threshold value.Accordingly, it clicks in each target user wait push
The probability of news is equal to or more than the target user of probability threshold value, the as potential user of the news to be pushed.Wherein, probability threshold
Value is set based on actual demand, it will be understood that when probability threshold value is set as lesser numerical value, can expand the potential of news to be pushed
User, and when probability threshold value is set as biggish numerical value, it can more precisely determine potential user.
For another example, for each news to be pushed, this can be clicked on wait push away to each target user according to numerical values recited
Each probability of news is sent to be ranked up, numerical value is sequentially reduced from front to back, and probability screening conditions may include clicking to be somebody's turn to do wait push away accordingly
Send the probabilistic of news in preceding predetermined figure.Wherein, predetermined figure can equal to or more than 1 positive integer in value, tool
Body can be preset according to actual needs.For example, predetermined figure is set as 10, then it is directed to each news to be pushed, by each target
10 target users of the maximum probability of the news to be pushed, the potential user as the news to be pushed are clicked in user.
S310 will respectively news to be pushed be associated with respective potential user.
News to be pushed is associated with potential user, can be and the news to be pushed is associated with pass with potential user foundation
System.Both it is associated with preservation with the user identifier of the potential user for example, the news of the news to be pushed can be identified, established with this
Between incidence relation.
In the present embodiment, for each news to be pushed, the news to be pushed and its each potential user are distinguished
Association.Aforementioned exemplary is accepted, as shown in Figure 6, it is assumed that from target user U1, U2, U3 and U4, determine news to be pushed
The potential user of News1 be target user U1 and U3, determine news News2 to be pushed potential user be target user U1,
U2 and U3 determines that the potential user of news News3 to be pushed is target user U2, U3 and U4.It then, will news be pushed
News1 is associated with target user U1 and is associated with news News1 to be pushed with target user U3;It similarly, will be new wait push
News2 is heard to be associated with target user U1, U2 and U3 respectively;Will news News3 be pushed respectively with target user U2, U3 and
U4 association.
S312 pushes news to each target user according to the associated news to be pushed of each target user.
It is appreciated that will be respectively after pushing news and being associated with respective potential user, wait push away associated by each target user
News is sent to can determine.Specifically, each target can be determined according to the association results of respectively news to be pushed and respective potential user
News to be pushed associated by user.
Aforementioned exemplary is accepted, news to be pushed associated by target user U1 is news News1 and News2 to be pushed, then
According to news News1 and News2 to be pushed, news is pushed to target user U1.News to be pushed associated by target user U2
News is pushed to target user U2 then according to news News2 and News3 to be pushed for news News2 and News3 to be pushed.Mesh
Marking news to be pushed associated by user U3 is news News1, News2 to be pushed and News3, then according to news to be pushed
News1, News2 and News3 push news to target user U3.
It specifically, can be associated respectively new wait push satisfaction in news by the target user for each target user
The news to be pushed for hearing screening conditions, is pushed to the target user.Wherein, news screening conditions can be set in advance according to actual needs
It is fixed.
It should be noted that can meet push trigger condition when, into associated new wait push according to each target user
Hear the step of pushing news to each target user (i.e. step S312).Correspondingly, it when being unsatisfactory for push trigger condition, does not then hold
Row step S312, and continue waiting for push trigger condition and meet.Wherein, push trigger condition can be set in advance according to actual needs
It is fixed, for example push trigger condition may include reaching at predetermined push time point, pushing trigger condition for another example may include receiving
From the push news acquisition request of target user.
In addition, pushing news to each target user is specifically to push news to user terminal locating for target user, due to
The mode that different user terminals receives news may be different, each target user can be directed to, according to the target user
The facility information of user terminal locating for associated news to be pushed and the target user pushes news to the user terminal.Its
In, facility information may include at least one in the system information for the operating system that device model information and equipment are run.
In practical applications, news push method can be applied to news application program, Tencent news client as shown in Figure 8
End.Specifically, as shown in figure 9, user's notification bar, subscriber terminal equipment can be transmitted in the news to be pushed for being pushed to target user
The related content for the news to be pushed that news push-delivery system is pushed can be shown on the display screen according to the display setting of itself
(such as title and abstract).
Above-mentioned news push method draws a portrait the user of each target user, inputs to have clicked the sample of news to be pushed
User's portrait of user be positive sample, sample of users to have received but not click on news to be pushed user to draw a portrait be negative sample
The click prediction model that model training obtains is carried out, then by clicking prediction model, is drawn a portrait according to the user of each target user pre-
The probability that each target user clicks on news to be pushed is surveyed, and then according to each probability, is determined from each target user wait push
The potential user of news.In this way, can determine that the potential use of news to be pushed without the news portrait for relying on news to be pushed
Family significantly reduces the dependence for the accuracy that the accuracy of news push draws a portrait to news.
In one embodiment, the mode of user's portrait of each target user is generated, it may include following steps: according to
The news category information for the news that each target user has clicked determines each target user respectively to the first of each predetermined news category
Preference weight;According to the news label that each news carries, determine that each target user is inclined to the second of each predetermined news label respectively
Good weight;According to the first preference weight corresponding with each target user and the second preference weight, generate each target user's
User's portrait.
First preference weight and predetermined news category can be one-to-one corresponding relationship.First preference weight can be used for table
Target user is levied to the preference of corresponding predetermined news category.
Second preference weight and predetermined news label can be one-to-one corresponding relationship.Second preference weight can be used for table
Target user is levied to the preference of corresponding predetermined news label.
The news category information of news, can be used for characterizing news category belonging to the news.Classification information can be by artificial true
Fixed (for example being determined by news editor personnel) can also be used any possible algorithm classification mode and determine.
The news label that news carries, can be used for describing the key content in the news.News label specifically can be newly
Keyword in news.News label can be can also be used any possible by manually determining (for example being determined by news editor personnel)
Tag extraction mode determines.
Specifically, for the target user that user's portrait is not present, under type such as can be used and generate user's portrait: obtaining and be somebody's turn to do
The news label that the news category information for the news that target user has clicked and the news clicked carry;Further according to having clicked
News news category information, determine the target user respectively to the first preference weight of each predetermined news label, and according to
The news label that each news clicked carries determines that the target user respectively weighs the second preference of each predetermined news label
Weight;In turn, according to the target user the first preference weight to each predetermined news label and respectively to each predetermined news mark respectively
Second preference weight of label generates user's portrait of the target user.Also that is, user's portrait of target user may include by the mesh
Mark user respectively the first preference weight to each predetermined news category and the target user respectively to each predetermined news label
Second preference weight combines the file to be formed.
It should be noted that target user can also have been obtained when determining the first preference weight and the second preference weight
The corresponding time decay factor of the news of click.It according to the news category information of the news clicked and can click new accordingly
Hear corresponding time decay factor, determine the target user respectively to the first preference weight of each predetermined news label, and according to
The news label and the corresponding time decay factor of news clicked that each news clicked carries, determine the target user point
Other the second preference weight to each predetermined news label.Wherein, the click time point of news is remoter apart from current point in time, this is new
Hear that corresponding time decay factor is bigger, otherwise the click time point of news is closer apart from current point in time, time decay factor
It is smaller.
It can be seen that time decay factor can be used for embodying news that target user has clicked to generating the target user's
The importance of user's portrait is gradually decreased with the time.Using time decay factor as factor, institute the considerations of generating user's portrait
User's portrait of generation can be conducive to the accuracy for improving news push closer to the actual conditions of target user.
It should be noted that can generate target user user portrait after store the user portrait, it is subsequent need using
When user's portrait of target user, the user that the target user is read from stored each user's portrait draws a portrait, and nothing
The news label of the interim news category information for executing the news clicked according to the target user and carrying is needed again to generate user
The step of portrait.
In addition, can be updated according to the scheduled update cycle to stored user portrait.Update cycle can be according to reality
Border demand is preset.For example, the update cycle is set as 1 day, then primary stored user's portrait is every other day updated.
Specifically, as shown in Figure 10, available when meeting portrait update condition (for example scheduled update time point reaches)
Terminal reported data, terminal reported data may include several user click datas, and each user click data may each comprise
Click mark (i.e. for characterizing the behavior mark that user behavior is click behavior), the user identifier of target user and the mesh
The news mark for the news item that mark user has clicked.In turn, portrait processing routine is called, for each user's hits
According to according to the news mark of news in the user click data, the news category information and the news for finding the news are carried
News label, the active user of the target user is found further according to the user identifier of target user in the user click data
Portrait, and then the news label carried according to the news category information of the news found and the news, update target use
The active user at family draws a portrait, and obtains updated user's portrait.News mark can be used for unique identification news, such as news ID.
Wherein, the news label carried according to the news category information of the news found and the news, updates the mesh
The active user's portrait for marking user, specifically can be by the active user of target user portrait with news belonging to the news
Corresponding current first preference weight of classification increases weight adjusted value, and will be new with this in the active user of target user portrait
It hears corresponding current second preference weight of the news label carried and increases weight adjusted value.Wherein, weight adjusted value is positive number, tool
Body can be preset according to actual needs.
For example, the active user of target user U1 draws a portrait, UP1 may include the target user to predetermined news category C1
Current first preference weight W11, to the current first preference weight W12 of predetermined news category C2, to predetermined news category C3
Current first preference weight W13, to the current first preference weight W14 of predetermined news category C4, to predetermined news label L1
Current second preference weight W21, to the current second preference weight W22 of predetermined news label L2 and to predetermined news mark
Sign the current second preference weight W23 of L3.
It is assumed that weight adjusted value is set as 1, a certain user click data CD1 includes clicking mark, the use of target user U1
The news mark for the news item that family mark and target user U1 have been clicked, and the news category letter of the news found
It is predetermined news label L3 that breath, which characterizes the news to belong to predetermined news category C2 and the news label of news carrying,.Accordingly,
It can be such that according to the concrete mode that user click data CD1 updates active user's portrait UP1 of target user U1 and use target
1 is increased to the current first preference weight W12 of predetermined news category C2 in active user's portrait UP1 of family U1, by active user
1 is increased to the current second preference weight W23 of predetermined news label L3 in portrait UP1.
In one embodiment, will the step of respectively news to be pushed is associated with respective potential user, i.e. step S310 can
Include the following steps: the user identifier of the respectively potential user of news to be pushed, corresponding write-in is right respectively with respectively news to be pushed
That answers respectively recalls in file.Accordingly, it according to the associated news to be pushed of each target user, is being pushed respectively to each target user new
Before the step of news, is i.e. before step S312, it may also include the steps of: and respectively called together according to corresponding with respectively news to be pushed
Back into file determines the associated news to be pushed of each target user respectively.
File is recalled, is the file for recording the user identifier of potential user of news to be pushed.Recall file with wait push
News can be one-to-one corresponding relationship.
In the present embodiment, for each news to be pushed, by the user identifier of the potential user of the news to be pushed,
News to be pushed is corresponding recalls file with this for write-in.Specifically, it can obtain and recall the corresponding file store path of file, in turn
According to this document store path, this is written in the user identifier of the potential user with the news to be pushed and is recalled in file.
For example, the potential user of news News1 to be pushed be target user U1 and U3, news News2's to be pushed
Potential user is target user U1, U2 and U3, and the potential user of news News3 to be pushed is target user U2, U3 and U4.
Then, by the user identifier of target user U1 and the user identifier of target user U3, write-in should news News1 be pushed be corresponding calls together
In back into file RF1;The user of the user identifier of target user U1, the user identifier of target user U2 and target user U3 is marked
Know, write-in should news News2 be pushed be corresponding recalls in file RF2;Also, by the user identifier of target user U2, target
The user identifier of user U3 and the user identifier of target user U4, write-in should news News3 be pushed be corresponding recalls file RF3
In.
In the present embodiment, can according to respectively news to be pushed is corresponding respectively recalls file, determine each target user
The news to be pushed being respectively associated.Aforementioned exemplary is accepted, can parse and recall file RF1, RF2 and RF3, so that it is determined that target out
The associated news to be pushed of user U1 be news News1 and News2 to be pushed, the associated news to be pushed of target user U2 be to
Push news News2 and News3, the associated news to be pushed of target user U3 be news News1, News2 to be pushed and
News3。
In one embodiment, according to the associated news to be pushed of each target user, news is pushed to each target user
Step, i.e. step S312, it may include following steps: associated according to the user of each target user portrait and each target user
The news of news to be pushed is drawn a portrait, and determines the recommendation index of the associated news to be pushed of each target user;It is closed from each target user
Connection wait push in news, filter out recommendation news corresponding with each target user;It is respectively right to each target user push
The recommendation news answered.
News portrait, can be used for characterizing the feature of news to be pushed.For example, news portrait may include the new of news to be pushed
It hears classification information and is somebody's turn to do the news label that news to be pushed carries.
The recommendation index of news to be pushed can be used for being characterized in each associated by the corresponding target user of the news to be pushed
Wait push in news, it is somebody's turn to do the recommendation degree of news to be pushed.In the present embodiment, for each target user, according to the mesh
Mark user user portrait and the associated news to be pushed of the target user news portrait, determine the target user it is associated to
Push the recommendation index of news.
By taking the associated news to be pushed of target user U3 is news News1, News2 to be pushed and News3 as an example, according to mesh
The news portrait for the user's portrait UP3 and news News1 to be pushed for marking user U3, determines that the recommendation of news News1 to be pushed refers to
Number, the recommendation index can be used for being characterized in news News1, News2 to be pushed and News3, and news News1 to be pushed is for mesh
Mark the recommendation degree of user U3;According to the news portrait of the user of target user U3 portrait UP3 and news News2 to be pushed, really
The recommendation index of fixed news News2 to be pushed, the recommendation index can be used for being characterized in news News1, News2 to be pushed and
In News3, recommendation degree of the news News2 to be pushed for target user U3;Also, it is drawn a portrait according to the user of target user U3
The news of UP3 and news News3 to be pushed portrait determine that the recommendation index of news News3 to be pushed, the recommendation index can be used for
It is characterized in news News1, News2 to be pushed and News3, recommendation degree of the news News3 to be pushed for target user U3.
Recommend news, it may include index is recommended to meet the news to be pushed of index screening conditions.Recommend news that is, needs
It is pushed to the news to be pushed of target user.Index screening conditions can be preset according to actual needs.
For example, index screening conditions may include being equal to or more than index threshold.It accordingly, should for each target user
Target user is associated respectively wait push in news, and index is recommended to be equal to or more than the news to be pushed of index threshold, the as mesh
Mark the corresponding recommendation news of user.Index threshold is set based on actual demand.
It for another example, can be associated to the target user respectively wait push according to numerical values recited for each target user
The recommendation index of news is ranked up, and numerical value is sequentially reduced from front to back, and index screening conditions may include recommending index row accordingly
It is listed in preceding predetermined figure.Wherein, predetermined figure can equal to or more than 1 positive integer in value, specifically can be according to practical need
It asks and presets.For example, predetermined figure is set as 5, it is for each target user, then the target user is associated respectively wait push
Recommend maximum 5 news to be pushed of index in news, as the corresponding recommendation news of the target user.
After determining the corresponding recommendation news of each target user, corresponding recommend newly can be pushed to each target user
It hears.For example, determining that target user U3 is corresponding from associated news News1, News2 to be pushed of target user U3 and News3
Recommendation news be news News1 and News2 to be pushed, then news News1 and News2 to be pushed is pushed to target user
U3。
In one embodiment, it draws a portrait by each user, inputs each click prediction corresponding with respectively news to be pushed
Before the step of model, i.e. before step S304, it may include following steps: according to the attribute type mark that respectively news to be pushed carries
Know, determines that respectively news to be pushed distinguishes matched push mode.Accordingly, step S304 may include following steps: each user is drawn
Picture, input belong in the corresponding each click prediction model of the news to be pushed of target push mode with each push mode.
Attribute type mark, is the information that can be used for characterizing the attribute type of news to be pushed, can be used as judging wait push away
Send the foundation of the matched push mode of news.For any news to be pushed, should matched push mode of news be pushed can be with
For target push mode or non-targeted push mode.Specifically, the attribute type of the attribute type mark characterization of news to be pushed
When for real time type or outburst type, it can determine that the matched push mode of news to be pushed is non-targeted push mode;Wait push away
When the attribute type of the attribute type mark characterization of news being sent to be the other types in addition to real time type and outburst type, it can sentence
Being somebody's turn to do the matched push mode of news to be pushed calmly is target push mode.In addition, the attribute type mark that news to be pushed carries
It can be by manually determining (for example being determined by news editor personnel).
It in the present embodiment, can be first according to the attribute type that respectively news to be pushed carries before executing step S304
Mark determines the matched push mode of news respectively to be pushed respectively.In turn, belong to target push mould for matched push mode
The news to be pushed of formula executes from the user of each target user draws a portrait to input and is somebody's turn to do the corresponding click prediction model of news to be pushed
To should news be pushed and the associated series of steps of its potential user, and then it is associated new wait push according to each target user
It hears to each target user and pushes news.Conversely, belonging to non-targeted push mode for matched push mode (is not belonging to target
Push mode) news to be pushed, then do not execute step S304 and subsequent step, and use other modes by the news to be pushed
It is pushed to respective objects user, for example directly gives the news push to be pushed to each target user, but not limited to this.
Belong to the news to be pushed of target push mode for matched push mode, just executes step S304 and subsequent step
Suddenly, can be effectively prevented from certain types of news to be pushed causes to fail normally to push due to using the push mode not being adapted to
The problem of, fail the problem of pushing in time such as what the news to be pushed of real time type and outburst type occurred.
In one embodiment, clicking prediction model may include XGBoost model.
XGBoost model, i.e. Extreme Gradient Boosting model, are Gradient Boosting
A C++ of Machine is realized.XGBoost model is a kind of promotion tree-model, is to integrate multiple tree-models to be formed
One strong classifier, tree-model can (Classfication And Regression Tree classifies and returns for CART tree-model
Gui Shu).
When carrying out model training, each positive sample sampled (sample of users of the news to be pushed can be clicked
User's portrait) and each negative sample user of the sample of users of news push (received but do not clicked on draw a portrait) with predetermined ratio
Randomly select out training set and test set.To which being obtained by training set training should the corresponding click prediction mould of news be pushed
Type.Also, by test set to news push corresponding click prediction model progress model performance evaluation.Wherein, make a reservation for
Ratio can be preset according to actual needs, for example can be set as 9:1, i.e. the number of samples of the number of samples of training set and test set
Ratio be 9:1, the clicks prediction model obtained under this ratio is with excellent performance.
For the optimization aim of model, AUC (Area Under Curve) this model-evaluation index can be selected.Wherein,
AUC is defined as the area of ROC (Receiver Operating Characteristic) curve following area, can be used for commenting
The superiority and inferiority of valence two-value classifier (Binary Classifier).
In a specific example, the highest depth of the tree in XGBoost model can be set as 5 layers, in training process
The number of iterations is 100 times.Accordingly, estimating the training time is 5 minutes, after training, can be stored in trained model local
Disk, for subsequent use.
Decision Tree algorithms are promoted it should be noted that carrying out model training to XGBoost model and can be related to gradient.It is right below
The basic principle of model training is briefly described:
Input: training set T={ (x1, y1), (x2, y2) ..., (xN, yN), (N indicates training by wherein i=1,2,3 ..., N
Concentrate the total number of sample).xiIndicate user's portrait of i-th of sample of users, yiIndicate standard results (the i.e. table of i-th of sample
Levy the probability that i-th of sample of users clicks accordingly news to be pushed, yiIt can be 1 or 0).
Output: regression tree fM(x)。
Wherein it is determined that regression tree fM(x) process can be related to following link (1) to (3):
(1) it initializes:
(2) to m successively value 1,2,3 ..., M (total number that M indicates tree), execute following steps 1. to step 4.:
Step is 1.: to i successively value be 1,2,3 ..., N, calculate
Step is 2.: regression tree is fitted to target rmi, estimate the leaf node region R of the m regression treemj, wherein j is indicated
Cutting variable, j=1,2,3 ..., J.That is, the m regression tree is a tree being made of J leaf node;
Step is 3.: to j successively value 1,2,3 ..., J, calculate
I.e. using the value in linear search estimation leaf node region, makes loss function minimization, find out optimal c in each leaf node region
Value;
Step is 4.: regression tree is updated,
(3) final regression tree is obtained:
The corresponding click prediction of news to be pushed is inputted to by user's portrait of target user with a specific example below
After model, the inner workings of model are briefly described.As shown in figure 11, it is somebody's turn to do the corresponding click prediction model of news to be pushed
Being integrated with M classification tree, (respectively Tree1 to TreeM), the user of the target user is drawn a portrait input should news pair be pushed
After the click prediction model answered, according to corresponding first preference weight of predetermined news category each in user portrait and each predetermined new
It hears corresponding second preference weight of label and since the root node of the classification tree, traverses its leaf section for every classification tree
Point until getting the corresponding prediction result of the classification tree, then will click on the corresponding prediction knot of each classification tree in prediction model
Fruit is cumulative, and obtaining the prediction result of the click prediction model, (target user that i.e. click prediction model is predicted clicks should be to
Push the probability of news).
It should be noted that by XGBoost model, can utilize automatically CPU (Central Processing Unit,
Central processing unit) multi-threading parallel process, while algorithmically being improved and improving precision.
In one embodiment, clicking prediction model may include FM model.
FM model, i.e. Factorization Machine model, also referred to as Factorization machine.FM model can be solved effectively
The certainly feature combinatorial problem in the case of Sparse.
In one embodiment, clicking prediction model may include FFM model.
FFM model, i.e. Field-aware Factorization Machine model, also referred to as field perceptual decomposition machine.FFM
Model is the upgrade version of FM model, and which introduce the concepts of Field, and the feature of same nature is incorporated into the same Field.
In one embodiment, clicking prediction model may include DeepFM model.
DeepFM model is the neural network model of integrated FM model and DNN model.DeepFM models coupling range and
The advantages of depth model, joint training FM model and DNN model, to learn the combination of low order feature and high-order feature group simultaneously
It closes.
In one embodiment, as shown in figure 12, a kind of news push method is provided.This method may include following steps
S1202 to S1210.
Historical user's behavioral data of each target user is sent to point by S1202, the corresponding user terminal of each target user
Cloth storage server cluster;Historical user's behavioral data of target user may include the new of the news that the target user has clicked
Hear ID.
S1204, distributed storage server cluster according to the news ID in historical user's behavioral data of each target user,
Determine the news category information of corresponding news and the news label that corresponding news carries.
S1206, distributed storage server cluster are true according to the news category information for the news that each target user has clicked
Fixed each target user is respectively to the first preference weight of each predetermined news category.
S1208, the news label that distributed storage server cluster is carried according to each news determine each target user's difference
To the second preference weight of each predetermined news label.
S1210, distributed storage server cluster are inclined to the first of each predetermined news category respectively according to each target user
Good weight and respectively to the second preference weight of each predetermined news label, generates user's portrait of each target user.
Respectively news to be pushed is sent to the corresponding user terminal of each sample of users by S1212, news the distribution server.
S1214, the corresponding user terminal of each sample of users will be used to characterize corresponding sample of users whether click to
The sample of users behavioral data for pushing news, is sent to distributed storage server cluster.
The sample of users behavioral data of each sample of users is sent to potential use by S1216, distributed storage server cluster
Family determines server.
S1218, potential user determine server for each news to be pushed, according to the sample of users of each sample of users
Behavioral data will click user's portrait of the sample of users of the news to be pushed as positive sample, and will receive but not
User's portrait of the sample of users of the news to be pushed is clicked as negative sample, model instruction is carried out according to positive sample and negative sample
Practice, obtains the corresponding click prediction model of the news to be pushed.
S1220, potential user determines that server obtains the news mark of news to be pushed, and is according to news mark judgement
It is no to there is click prediction model corresponding with news to be pushed.If it exists, then S1222 is jumped to;If it does not exist, then without pushing away
It send, and continues waiting for, and be recorded as " there is no click prediction model " not push reason (not shown).
S1222, distributed storage server cluster draw a portrait the user of each target user, are sent to potential user and determine clothes
Business device.
S1224, loads stored click prediction model corresponding with news to be pushed, and by the user of each target user
Portrait inputs click prediction model corresponding with respectively news to be pushed.
S1226, potential user determine that server by each click prediction model, is drawn a portrait according to the user of each target user,
Predict that each target user clicks on the probability of respectively news to be pushed.
S1228, potential user determine server according to each probability, the latent of news respectively to be pushed are determined from each target user
In user, and by the user identifier of the respectively potential user of news to be pushed, corresponding write-in is corresponding with respectively news to be pushed
Respectively recall in file.
S1230, potential user determine that server corresponding with respectively news to be pushed will respectively recall file and be sent to newly
Hear evaluating server.
S1232, news evaluating server respectively recall file according to corresponding with respectively news to be pushed, and determine respectively each
The associated news to be pushed of target user, and it is associated new wait push according to the user of each target user portrait and each target user
The news of news is drawn a portrait, and determines the recommendation index for the news to be pushed that each target user is respectively associated.
The recommendation index of the associated news to be pushed of each target user is sent to news by S1234, news evaluating server
The distribution server.
S1236, news the distribution server are directed to each target user, associated respectively new wait push according to the target user
The recommendation index of news, it is associated respectively wait push in news from the target user, filter out the corresponding recommendation news of the target user.
The corresponding recommendation news of each target user is corresponded to and is pushed to each target user point by S1238, news the distribution server
Not corresponding user terminal.
It should be noted that the specific restriction of each technical characteristic in the present embodiment, can with hereinbefore to relevant art
The restriction of feature is identical, is not added and repeats herein.
In one embodiment, as shown in figure 13, the determination method of the potential user of news a kind of is provided.This method can
To be executed as above-mentioned Fig. 1 and news push-delivery system shown in Fig. 2 120, (for shown in Fig. 2, this method can specifically be pushed away by news
The potential user in system 120 is sent to determine that server 1204 executes).This method may include following steps S1302 to S1308.
S1302 obtains user's portrait of each target user.
S1304 draws a portrait each user, inputs click prediction model corresponding with news to be pushed;Click prediction model with
The user's portrait for having clicked the sample of users of news to be pushed is positive sample, the sample to have received but not click on news to be pushed
User's portrait of user is that negative sample progress model training obtains;The number of news to be pushed can be equal to or be greater than 1.
S1306 draws a portrait according to the user of each target user by clicking prediction model, predicts each target user point respectively
Hit the probability of news to be pushed.
S1308 determines the potential user of news to be pushed according to each probability from each target user.
The determination method of the potential user of above-mentioned news draws a portrait the user of each target user, inputs to have clicked wait push away
Send the user of sample of users of the user's portrait of the sample of users of news for positive sample, to have received but not click on news to be pushed
Portrait is the click prediction model that negative sample carries out that model training obtains, then by clicking prediction model, according to each target user
User draw a portrait the probability for predicting that each target user clicks on news to be pushed, and then according to each probability, from each target user
The potential user of middle determination news to be pushed.In this way, can determine that without the news portrait for relying on news to be pushed wait push
The potential user of news can be effectively reduced the dependence for the accuracy that the accuracy of news push draws a portrait to news.
In one embodiment, the mode of user's portrait of each target user is generated, it may include following steps: according to each mesh
The news category information for the news that mark user has clicked, determines each target user respectively to the first preference of each predetermined news category
Weight;According to the news label that each news carries, determine that each target user respectively weighs the second preference of each predetermined news label
Weight;According to the first preference weight corresponding with each target user and the second preference weight, the user of each target user is generated
Portrait.
In one embodiment, according to each probability, determine the potential user's of news to be pushed from each target user
After step, is i.e. after step S1308, it may further comprise the step of: and be associated with news to be pushed with its potential user.
It in one embodiment, will news be pushed the step of being associated with its potential user, it may include following steps: will be to
The user identifier for pushing the potential user of news, is written that news to be pushed is corresponding to recall file.
In one embodiment, clicking prediction model includes XGBoost model, FM model, FFM model and DeepFM
Any one in model.
In one embodiment, the push of news to be pushed is determined in the attribute type mark for including according to news to be pushed
It when mode belongs to target push mode, draws a portrait by each user, inputs click prediction model corresponding with news to be pushed
Step.
It should be noted that special to each technology in the determination method of the potential user for the news that each embodiment of the application provides
Sign specific restriction, can with it is hereinbefore identical to the restriction of relevant art feature in news push method, be not added and repeat herein.
It should be appreciated that although each step in the flow chart that each embodiment is related to above is according to arrow under reasonable terms
Instruction successively show that but these steps are not that the inevitable sequence according to arrow instruction successively executes.Unless having herein
Explicitly stated, there is no stringent sequences to limit for the execution of these steps, these steps can execute in other order.And
And at least part step in each flow chart may include multiple sub-steps perhaps these sub-steps of multiple stages or stage
It is not necessarily to execute completion in synchronization, but can execute at different times, these sub-steps or stage hold
Row sequence is also not necessarily and successively carries out, but can be with the sub-step or stage of other steps or other steps at least
A part executes in turn or alternately.
In one embodiment, as shown in figure 14, a kind of news push device 1400 is provided.The device 1400 may include
Following module 1402 to 1412.
First user, which draws a portrait, obtains module 1402, and the user for obtaining each target user draws a portrait.
First portrait input module 1404 inputs corresponding with respectively news to be pushed each for each user to draw a portrait
Click prediction model;Wherein, news to be pushed is corresponding clicks prediction model to have clicked the sample of users of the news to be pushed
User's portrait be positive sample, sample of users to have received but not click on the news to be pushed user draw a portrait be negative sample into
Row model training obtains.
First probability determination module 1406, for being drawn a portrait according to the user of each target user by each click prediction model,
Predict that each target user clicks on the probability of respectively news to be pushed.
First potential user's determining module 1408 is respectively new wait push for being determined from each target user according to each probability
The potential user of news.
First relating module 1410, for will respectively news to be pushed be associated with respective potential user.
News push module 1412, for being pushed to each target user according to the associated news to be pushed of each target user
News.
Above-mentioned news push device 1400 draws a portrait the user of each target user, inputs to have clicked news to be pushed
User's portrait of sample of users of the user's portrait of sample of users for positive sample, to have received but not click on news to be pushed is negative
Sample carries out the click prediction model that model training obtains, then by clicking prediction model, is drawn according to the user of each target user
As predicting that each target user clicks on the probability of news to be pushed, and then according to each probability, determined from each target user to
Push the potential user of news.In this way, can determine that the latent of news to be pushed without the news portrait for relying on news to be pushed
In user, the dependence for the accuracy that the accuracy of news push draws a portrait to news is significantly reduced.
In one embodiment, news push device 1400 may also include following module: the first preference weight determining module,
The news category information of news for having been clicked according to each target user determines each target user respectively to each predetermined news category
Other first preference weight;Second preference weight determining module, the news label for being carried according to each news, determines each target
User is respectively to the second preference weight of each predetermined news label;First user portrait generation module, for basis and each target
Corresponding first preference weight of user and the second preference weight generate user's portrait of each target user.
In one embodiment, the first relating module 1402 may include such as lower unit: the first user identifier writing unit, use
In by the user identifier of the respectively potential user of news to be pushed, corresponding write-in with respectively news to be pushed is corresponding respectively recalls text
In part.Accordingly, news push device 1400 may also include association news determining module, for distinguishing according to respectively news to be pushed
It is corresponding respectively to recall file, the associated news to be pushed of each target user is determined respectively.
In one embodiment, news push module 1402 may include such as lower unit: recommending index determination unit, is used for root
According to user's portrait of each target user and the news portrait of the associated news to be pushed of each target user, determine that each target is used
The recommendation index of the associated news to be pushed in family;Recommend news determination unit, for associated new wait push from each target user
Wen Zhong filters out recommendation news corresponding with each target user;Recommending news includes that index is recommended to meet index screening item
The news to be pushed of part;Recommend news push module, for pushing corresponding recommendation news to each target user.
In one embodiment, click prediction model may include XGBoost model, FM model, FFM model and
Any one in DeepFM model.
In one embodiment, news push device 1400 may also include push mode determining module, for according to respectively to
The attribute type mark that news carries is pushed, determines that respectively news to be pushed distinguishes matched push mode.Accordingly, the first portrait is defeated
Entering module 1402 can be used for drawing a portrait each user, and the news to be pushed that input belongs to target push mode with each push mode is distinguished
In corresponding each click prediction model.
It should be noted that the specific restriction about news push device 1400, may refer to push away above for news
The restriction of delivery method, details are not described herein.Modules in above-mentioned news push device 1400 can be fully or partially through soft
Part, hardware and combinations thereof are realized.Above-mentioned each module can be embedded in the form of hardware or independently of the processing in computer equipment
It in device, can also be stored in a software form in the memory in computer equipment, in order to which processor calls execution above each
The corresponding operation of a module.
In one embodiment, as shown in figure 15, the determining device 1500 of the potential user of news a kind of is provided.The dress
Setting 1500 may include following module 1502 to 1508.
Second user portrait obtains module 1502, and the user for obtaining each target user draws a portrait.
Second portrait input module 1504 inputs click prediction corresponding with news to be pushed for each user to draw a portrait
Model;Click prediction model and be with the user's portrait for having clicked the sample of users of news to be pushed positive sample, to have received but
The user's portrait for not clicking on the sample of users of news to be pushed is that negative sample progress model training obtains.
Second probability determination module 1506, for being drawn a portrait according to the user of each target user, in advance by clicking prediction model
Survey the probability that each target user clicks on news to be pushed.
Second potential user's determining module 1508, for determining news to be pushed from each target user according to each probability
Potential user.
The determining device 1500 of the potential user of above-mentioned news draws a portrait the user of each target user, inputs to have clicked
User's portrait of the sample of users of news to be pushed is positive sample, the sample of users to have received but not click on news to be pushed
User's portrait is the click prediction model that negative sample carries out that model training obtains, then by clicking prediction model, according to each target
User's portrait of user predicts that each target user clicks on the probability of news to be pushed, and then according to each probability, from each target
The potential user of news to be pushed is determined in user.In this way, without rely on news to be pushed news portrait, that is, can determine that
The potential user for pushing news can be effectively reduced the dependence for the accuracy that the accuracy of news push draws a portrait to news.
In one embodiment, the determining device 1500 of the potential user of news may also include following module: third preference
Weight determination module, the news category information of the news for having been clicked according to each target user determine each target user's difference
To the first preference weight of each predetermined news category;4th preference weight determining module, the news for being carried according to each news
Label determines each target user respectively to the second preference weight of each predetermined news label;Second user portrait generation module, is used
According to the first preference weight corresponding with each target user and the second preference weight, the user for generating each target user is drawn
Picture.
In one embodiment, the determining device 1500 of the potential user of news may also include the second relating module, be used for
News to be pushed is associated with its potential user.
In one embodiment, the second relating module may include second user mark writing unit, for will be new wait push
The user identifier of the potential user of news, is written that news to be pushed is corresponding to recall file.
In one embodiment, click prediction model may include XGBoost model, FM model, FFM model and
Any one in DeepFM model.
In one embodiment, the determining device 1500 of the potential user of news may also include trigger module, in root
When according to belonging to target push mode wait push the push mode of news wait push the attribute type mark determination that news includes, triggering
The second portrait input module 1504 is called, each user is drawn a portrait with entering, click corresponding with news to be pushed is inputted and predicts mould
The step of type.
It should be noted that the specific restriction of the determining device 1500 of the potential user about news, may refer to above
In for news potential user determination method restriction, details are not described herein.The potential user's of above-mentioned news determines dress
Setting the modules in 1500 can realize fully or partially through software, hardware and combinations thereof.Above-mentioned each module can be with hardware
Form is embedded in or independently of in the processor in computer equipment, can also be stored in computer equipment in a software form
In memory, the corresponding operation of the above modules is executed in order to which processor calls.
In one embodiment, a kind of computer equipment, including memory and processor are provided, memory is stored with meter
Calculation machine program, when computer program is executed by processor, so that processor executes the latent of above-mentioned news push method and/or news
Step in the determination method of user.The news that the step in news push method can be above-mentioned each embodiment herein pushes away
Step in delivery method, similarly, the step in the determination method of the potential user of news can be above-mentioned each embodiment
Step in the determination method of the potential user of news.
The internal structure chart of computer equipment in one embodiment is shown in Figure 16, which includes passing through to be
Processor, memory, network interface and the database of bus of uniting connection.Wherein, which calculates and controls energy for providing
Power.The memory includes non-volatile memory medium and built-in storage, which is stored with operating system, meter
Calculation machine program and database, the built-in storage are that the operation of the operating system and computer program in non-volatile memory medium mentions
For environment.The database is used to store the data such as user's portrait of target user.The network interface is used for logical with external terminal
Cross network connection communication.To realize the latent of above-mentioned news push method and/or news when the computer program is executed by processor
Step in the determination method of user.
It will be understood by those skilled in the art that structure shown in Figure 16, only part relevant to application scheme
The block diagram of structure, does not constitute the restriction for the computer equipment being applied thereon to application scheme, and specific computer is set
Standby may include perhaps combining certain components or with different component layouts than more or fewer components as shown in the figure.
In one embodiment, news push device 1400 provided by the present application can be implemented as a kind of computer program
Form, computer program can be run in computer equipment as shown in figure 16.Group can be stored in the memory of computer equipment
At each program module of the news push device 1400, for example the first user shown in Figure 14 draws a portrait and obtains module 1402, the
One portrait input module 1404, the first probability determination module 1406 and first potential user's determining module 1408 etc..Each journey
The news that the computer program of sequence module composition makes processor execute each embodiment of the application described in this specification pushes away
Step in delivery method.
For example, computer equipment shown in Figure 16 can pass through first in news push device 1400 as shown in figure 14
User draws a portrait and obtains the execution of module 1402 step S302, executes step S304 etc. by the first portrait input module 1404.
In one embodiment, the determining device 1500 of the potential user of news provided by the present application can be implemented as one kind
The form of computer program, computer program can be run in computer equipment as shown in figure 16.The storage of computer equipment
The each program module for forming the determining device 1500 of potential user of the news, such as shown in figure 15 second can be stored in device
User, which draws a portrait, obtains the portrait of module 1502, second input module 1504, the second probability determination module 1506 and the second potential use
Family determining module 1508.The computer program that each program module is constituted makes processor execute this Shen described in this specification
It please step in the news push method of each embodiment.
For example, computer equipment shown in Figure 16 can pass through the determining device of the potential user of the news as shown in figure R
Second user portrait in 1500 obtains module 1502 and executes step S1302, executes step by the second portrait input module 1504
Rapid S1304 etc..
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a non-volatile computer and can be read
In storage medium, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, provided herein
Each embodiment used in any reference to memory, storage, database or other media, may each comprise non-volatile
And/or volatile memory.Nonvolatile memory may include that read-only memory (ROM), programming ROM (PROM), electricity can be compiled
Journey ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include random access memory
(RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms, such as static state RAM
(SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhanced SDRAM
(ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) directly RAM (RDRAM), straight
Connect memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
Accordingly, in one embodiment, a kind of computer readable storage medium is provided, computer program is stored with, is counted
When calculation machine program is executed by processor, so that processor executes the potential user of above-mentioned news push method and/or news really
Determine the step in method.The step in news push method can be in the news push method of above-mentioned each embodiment herein
Step, similarly, the step in the determination method of the potential user of news can be the potential of the news of above-mentioned each embodiment
Step in the determination method of user.
Each technical characteristic of above embodiments can be combined arbitrarily, for simplicity of description, not to above-described embodiment
In each technical characteristic it is all possible combination be all described, as long as however, the combination of these technical characteristics be not present lance
Shield all should be considered as described in this specification.
The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
The limitation to the application the scope of the patents therefore cannot be interpreted as.It should be pointed out that for those of ordinary skill in the art
For, without departing from the concept of this application, various modifications and improvements can be made, these belong to the guarantor of the application
Protect range.Therefore, the scope of protection shall be subject to the appended claims for the application patent.
Claims (10)
1. a kind of news push method, comprising:
Obtain user's portrait of each target user;
Each user is drawn a portrait, each click prediction model corresponding with respectively news to be pushed is inputted;Wherein, described wait push away
Send news is corresponding to click prediction model to have clicked the user of the sample of users of the news to be pushed and draw a portrait for positive sample, with
The user's portrait for receiving but not clicking on the sample of users of the news to be pushed is that negative sample progress model training obtains;
It by each click prediction model, is drawn a portrait according to the user of each target user, predicts each target user point
The probability of respectively news to be pushed is not clicked;
According to each probability, the potential user of each news to be pushed is determined from each target user;
Each news to be pushed is associated with respective potential user;
According to each associated news to be pushed of target user, news is pushed to each target user.
2. the method according to claim 1, wherein generate the mode of user's portrait of each target user,
Include:
According to the news category information for the news that each target user has clicked, determine each target user respectively to each pre-
Determine the first preference weight of news category;
According to the news label that each news carries, determine each target user respectively to the second of each predetermined news label
Preference weight;
According to the first preference weight corresponding with each target user and the second preference weight, generates each target and use
The user at family draws a portrait.
3. the method according to claim 1, wherein described by each news to be pushed and respective potential use
Family association, comprising:
By the user identifier of the potential user of each news to be pushed, corresponding write-in is respectively corresponded with each news to be pushed
Respectively recall in file;
Described respectively according to each associated news to be pushed of target user, to each target user push news it
Before, further includes:
Respectively recall file according to corresponding with each news to be pushed, determine respectively each target user it is associated to
Push news.
4. the method according to claim 1, wherein described associated new wait push according to each target user
It hears, pushes news to each target user, comprising:
It is drawn according to the news of the user of each target user portrait and the associated news to be pushed of each target user
Picture determines the recommendation index of the associated news to be pushed of each target user;
It is associated wait push in news from each target user, filter out recommend newly corresponding with each target user
It hears;The recommendation news includes the news to be pushed for recommending index to meet index screening conditions;
Corresponding recommendation news is pushed to each target user.
5. the method according to claim 1, wherein the click prediction model includes XGBoost model, FM mould
Any one in type, FFM model and DeepFM model.
6. the method according to any one of claims 1 to 5, which is characterized in that each user is drawn a portrait described, it is defeated
Before entering each click prediction model corresponding with respectively news to be pushed, comprising:
According to the attribute type mark that each news to be pushed carries, determine that each news to be pushed distinguishes matched push
Mode;
It is described that each user draws a portrait, input each click prediction model corresponding with respectively news to be pushed, comprising:
Each user is drawn a portrait, it is corresponding respectively that input with each push mode belongs to the news to be pushed of target push mode
It clicks in prediction model.
7. a kind of determination method of the potential user of news, comprising:
Obtain user's portrait of each target user;
Each user is drawn a portrait, click prediction model corresponding with news to be pushed is inputted;The click prediction model be with
It is positive sample that the user for having clicked the sample of users of the news to be pushed, which draws a portrait, described new wait push to have received but not click on
User's portrait of the sample of users of news is that negative sample progress model training obtains;
It by the click prediction model, is drawn a portrait according to the user of each target user, predicts each target user's difference
Click the probability of the news to be pushed;
According to each probability, the potential user of the news to be pushed is determined from each target user.
8. a kind of news push device, comprising:
First user, which draws a portrait, obtains module, and the user for obtaining each target user draws a portrait;
First portrait input module inputs each click corresponding with respectively news to be pushed for each user to draw a portrait
Prediction model;Wherein, the corresponding prediction model of clicking of the news to be pushed is to have clicked the sample of the news to be pushed and use
User's portrait at family be positive sample, sample of users to have received but not click on the news to be pushed user to draw a portrait be negative sample
Model training is carried out to obtain;
First probability determination module, for being drawn a portrait according to the user of each target user by each click prediction model,
Predict that each target user clicks on the probability of respectively news to be pushed;
First potential user's determining module, it is each described wait push away for being determined from each target user according to each probability
Send the potential user of news;
First relating module, for each news to be pushed to be associated with respective potential user;
News push module, for being pushed to each target user according to each associated news to be pushed of target user
News.
9. a kind of determining device of the potential user of news, comprising:
Second user portrait obtains module, and the user for obtaining each target user draws a portrait;
Second portrait input module inputs click prediction model corresponding with news to be pushed for each user to draw a portrait;
The prediction model of clicking is to be drawn a portrait with the user for having clicked the sample of users of the news to be pushed for positive sample, to have received
But the user's portrait for not clicking on the sample of users of the news to be pushed is that negative sample progress model training obtains;
Second probability determination module, for being drawn a portrait according to the user of each target user, in advance by the click prediction model
Survey the probability that each target user clicks on the news to be pushed;
Second potential user's determining module, it is described wait push for being determined from each target user according to each probability
The potential user of news.
10. a kind of computer equipment, including memory and processor, the memory is stored with computer program, the calculating
When machine program is executed by the processor, so that the processor executes the step such as any one of claims 1 to 7 the method
Suddenly.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811601767.4A CN109684554B (en) | 2018-12-26 | 2018-12-26 | Method for determining potential users of news and news pushing method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811601767.4A CN109684554B (en) | 2018-12-26 | 2018-12-26 | Method for determining potential users of news and news pushing method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109684554A true CN109684554A (en) | 2019-04-26 |
CN109684554B CN109684554B (en) | 2021-07-20 |
Family
ID=66189756
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811601767.4A Active CN109684554B (en) | 2018-12-26 | 2018-12-26 | Method for determining potential users of news and news pushing method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109684554B (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110363427A (en) * | 2019-07-15 | 2019-10-22 | 腾讯科技(深圳)有限公司 | Model quality evaluation method and apparatus |
CN110659414A (en) * | 2019-09-02 | 2020-01-07 | 惠州学院 | User portrait based personalized information pushing method and user side |
CN110688553A (en) * | 2019-08-13 | 2020-01-14 | 平安科技(深圳)有限公司 | Information pushing method and device based on data analysis, computer equipment and storage medium |
CN111259256A (en) * | 2020-02-13 | 2020-06-09 | 腾讯科技(深圳)有限公司 | Content processing method, content processing device, computer-readable storage medium and computer equipment |
CN111523027A (en) * | 2020-04-16 | 2020-08-11 | 武汉有牛科技有限公司 | Automatic data news writing robot based on block chain technology |
CN111582932A (en) * | 2020-03-25 | 2020-08-25 | 平安壹钱包电子商务有限公司 | Inter-scene information pushing method and device, computer equipment and storage medium |
CN111666252A (en) * | 2020-05-27 | 2020-09-15 | 上海连尚网络科技有限公司 | Method and device for obtaining recommendation popularity information of recommended books |
CN112070532A (en) * | 2020-08-27 | 2020-12-11 | 中国平安财产保险股份有限公司 | Information pushing method, device, equipment and storage medium |
CN112131456A (en) * | 2019-06-24 | 2020-12-25 | 腾讯科技(北京)有限公司 | Information pushing method, device, equipment and storage medium |
CN112148994A (en) * | 2020-10-23 | 2020-12-29 | 腾讯科技(深圳)有限公司 | Information push effect evaluation method and device, electronic equipment and storage medium |
CN112508613A (en) * | 2020-12-09 | 2021-03-16 | 天津狮拓信息技术有限公司 | Commodity recommendation method and device, electronic equipment and readable storage medium |
CN112925972A (en) * | 2019-12-06 | 2021-06-08 | 北京达佳互联信息技术有限公司 | Information pushing method and device, electronic equipment and storage medium |
CN112925982A (en) * | 2021-03-12 | 2021-06-08 | 上海意略明数字科技股份有限公司 | User redirection method and device, storage medium and computer equipment |
CN113378067A (en) * | 2021-07-13 | 2021-09-10 | 深圳前海微众银行股份有限公司 | Message recommendation method, device, medium, and program product based on user mining |
CN113543117A (en) * | 2020-04-22 | 2021-10-22 | ***通信集团重庆有限公司 | Prediction method and device for number portability user and computing equipment |
CN113722498A (en) * | 2021-06-17 | 2021-11-30 | 武汉理工数字传播工程有限公司 | Method, device, equipment and storage medium for digital resource management |
CN115080811A (en) * | 2022-06-17 | 2022-09-20 | 抖音视界(北京)有限公司 | Information recommendation method, search recommendation word display method and device |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2533430A2 (en) * | 2010-02-03 | 2012-12-12 | Korea University Research And Business Foundation | Portable communication terminal for extracting subjects of interest to the user, and a method therefor |
CN105005587A (en) * | 2015-06-26 | 2015-10-28 | 深圳市腾讯计算机***有限公司 | User portrait updating method, apparatus and system |
CN105608171A (en) * | 2015-12-22 | 2016-05-25 | 青岛海贝易通信息技术有限公司 | User portrait construction method |
US20160179968A1 (en) * | 2014-12-22 | 2016-06-23 | Facebook, Inc. | Systems and methods for recommending follow up content |
CN107341272A (en) * | 2017-08-25 | 2017-11-10 | 北京奇艺世纪科技有限公司 | A kind of method for pushing, device and electronic equipment |
CN107766580A (en) * | 2017-11-20 | 2018-03-06 | 北京奇虎科技有限公司 | The method for pushing and device of message |
CN108521439A (en) * | 2018-03-08 | 2018-09-11 | 北京三快在线科技有限公司 | A kind of method and apparatus of message push |
CN108769159A (en) * | 2018-05-16 | 2018-11-06 | 北京豆果信息技术有限公司 | A kind of electronic cookbook intelligent recommendation method |
-
2018
- 2018-12-26 CN CN201811601767.4A patent/CN109684554B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2533430A2 (en) * | 2010-02-03 | 2012-12-12 | Korea University Research And Business Foundation | Portable communication terminal for extracting subjects of interest to the user, and a method therefor |
US20160179968A1 (en) * | 2014-12-22 | 2016-06-23 | Facebook, Inc. | Systems and methods for recommending follow up content |
CN105005587A (en) * | 2015-06-26 | 2015-10-28 | 深圳市腾讯计算机***有限公司 | User portrait updating method, apparatus and system |
CN105608171A (en) * | 2015-12-22 | 2016-05-25 | 青岛海贝易通信息技术有限公司 | User portrait construction method |
CN107341272A (en) * | 2017-08-25 | 2017-11-10 | 北京奇艺世纪科技有限公司 | A kind of method for pushing, device and electronic equipment |
CN107766580A (en) * | 2017-11-20 | 2018-03-06 | 北京奇虎科技有限公司 | The method for pushing and device of message |
CN108521439A (en) * | 2018-03-08 | 2018-09-11 | 北京三快在线科技有限公司 | A kind of method and apparatus of message push |
CN108769159A (en) * | 2018-05-16 | 2018-11-06 | 北京豆果信息技术有限公司 | A kind of electronic cookbook intelligent recommendation method |
Non-Patent Citations (1)
Title |
---|
匡文波: "论视频智能推荐的算法模型设计", 《新闻传播》 * |
Cited By (26)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112131456A (en) * | 2019-06-24 | 2020-12-25 | 腾讯科技(北京)有限公司 | Information pushing method, device, equipment and storage medium |
CN110363427A (en) * | 2019-07-15 | 2019-10-22 | 腾讯科技(深圳)有限公司 | Model quality evaluation method and apparatus |
CN110688553A (en) * | 2019-08-13 | 2020-01-14 | 平安科技(深圳)有限公司 | Information pushing method and device based on data analysis, computer equipment and storage medium |
WO2021027362A1 (en) * | 2019-08-13 | 2021-02-18 | 平安科技(深圳)有限公司 | Information pushing method and apparatus based on data analysis, computer device, and storage medium |
CN110659414A (en) * | 2019-09-02 | 2020-01-07 | 惠州学院 | User portrait based personalized information pushing method and user side |
CN110659414B (en) * | 2019-09-02 | 2023-01-13 | 惠州学院 | User portrait based personalized information pushing method and user side |
CN112925972A (en) * | 2019-12-06 | 2021-06-08 | 北京达佳互联信息技术有限公司 | Information pushing method and device, electronic equipment and storage medium |
CN112925972B (en) * | 2019-12-06 | 2024-03-08 | 北京达佳互联信息技术有限公司 | Information pushing method, device, electronic equipment and storage medium |
CN111259256A (en) * | 2020-02-13 | 2020-06-09 | 腾讯科技(深圳)有限公司 | Content processing method, content processing device, computer-readable storage medium and computer equipment |
CN111259256B (en) * | 2020-02-13 | 2023-09-12 | 深圳市雅阅科技有限公司 | Content processing method, content processing device, computer readable storage medium and computer equipment |
CN111582932A (en) * | 2020-03-25 | 2020-08-25 | 平安壹钱包电子商务有限公司 | Inter-scene information pushing method and device, computer equipment and storage medium |
CN111523027B (en) * | 2020-04-16 | 2023-08-01 | 武汉有牛科技有限公司 | Automatic data news writing robot based on blockchain technology |
CN111523027A (en) * | 2020-04-16 | 2020-08-11 | 武汉有牛科技有限公司 | Automatic data news writing robot based on block chain technology |
CN113543117B (en) * | 2020-04-22 | 2022-10-04 | ***通信集团重庆有限公司 | Prediction method and device for number portability user and computing equipment |
CN113543117A (en) * | 2020-04-22 | 2021-10-22 | ***通信集团重庆有限公司 | Prediction method and device for number portability user and computing equipment |
CN111666252B (en) * | 2020-05-27 | 2023-09-15 | 上海连尚网络科技有限公司 | Method and equipment for acquiring recommended heat information of recommended books |
CN111666252A (en) * | 2020-05-27 | 2020-09-15 | 上海连尚网络科技有限公司 | Method and device for obtaining recommendation popularity information of recommended books |
CN112070532A (en) * | 2020-08-27 | 2020-12-11 | 中国平安财产保险股份有限公司 | Information pushing method, device, equipment and storage medium |
CN112148994A (en) * | 2020-10-23 | 2020-12-29 | 腾讯科技(深圳)有限公司 | Information push effect evaluation method and device, electronic equipment and storage medium |
CN112148994B (en) * | 2020-10-23 | 2024-04-19 | 腾讯科技(深圳)有限公司 | Information push effect evaluation method and device, electronic equipment and storage medium |
CN112508613A (en) * | 2020-12-09 | 2021-03-16 | 天津狮拓信息技术有限公司 | Commodity recommendation method and device, electronic equipment and readable storage medium |
CN112508613B (en) * | 2020-12-09 | 2024-03-19 | 天津狮拓信息技术有限公司 | Commodity recommendation method and device, electronic equipment and readable storage medium |
CN112925982A (en) * | 2021-03-12 | 2021-06-08 | 上海意略明数字科技股份有限公司 | User redirection method and device, storage medium and computer equipment |
CN113722498A (en) * | 2021-06-17 | 2021-11-30 | 武汉理工数字传播工程有限公司 | Method, device, equipment and storage medium for digital resource management |
CN113378067A (en) * | 2021-07-13 | 2021-09-10 | 深圳前海微众银行股份有限公司 | Message recommendation method, device, medium, and program product based on user mining |
CN115080811A (en) * | 2022-06-17 | 2022-09-20 | 抖音视界(北京)有限公司 | Information recommendation method, search recommendation word display method and device |
Also Published As
Publication number | Publication date |
---|---|
CN109684554B (en) | 2021-07-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109684554A (en) | The determination method and news push method of the potential user of news | |
US11983269B2 (en) | Deep neural network system for similarity-based graph representations | |
CN108304379B (en) | Article identification method and device and storage medium | |
CN110458220B (en) | Crowd orientation method, device, server and storage medium | |
CN110457581A (en) | A kind of information recommended method, device, electronic equipment and storage medium | |
CN108885624B (en) | Information recommendation system and method | |
US8082247B2 (en) | Best-bet recommendations | |
CN107341187B (en) | Search processing method, device, equipment and computer storage medium | |
CN110968695A (en) | Intelligent labeling method, device and platform based on active learning of weak supervision technology | |
CN109446407A (en) | Correlation recommendation method, apparatus, computer equipment and storage medium | |
CN110888990A (en) | Text recommendation method, device, equipment and medium | |
KR20160058896A (en) | System and method for analyzing and transmitting social communication data | |
CN104899315A (en) | Method and device for pushing user information | |
CN104217030A (en) | Method and device for classifying users according to search log data of server | |
CN112183994A (en) | Method and device for evaluating equipment state, computer equipment and storage medium | |
KR101725510B1 (en) | Method and apparatus for recommendation of social event based on users preference | |
CN110175264A (en) | Construction method, server and the computer readable storage medium of video user portrait | |
CN106095738A (en) | Recommendation tables single slice | |
CN107291774B (en) | Error sample identification method and device | |
CN109271624A (en) | A kind of target word determines method, apparatus and storage medium | |
CN110110218A (en) | A kind of Identity Association method and terminal | |
CN111582448B (en) | Weight training method and device, computer equipment and storage medium | |
CN114443671A (en) | Recommendation model updating method and device, computer equipment and storage medium | |
CN104572887A (en) | Method and system for retrieving product information | |
CN112330387A (en) | Virtual broker applied to house-watching software |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20221122 Address after: 1402, Floor 14, Block A, Haina Baichuan Headquarters Building, No. 6, Baoxing Road, Haibin Community, Xin'an Street, Bao'an District, Shenzhen, Guangdong 518100 Patentee after: Shenzhen Yayue Technology Co.,Ltd. Address before: 518000 Tencent Building, No. 1 High-tech Zone, Nanshan District, Shenzhen City, Guangdong Province, 35 Floors Patentee before: TENCENT TECHNOLOGY (SHENZHEN) Co.,Ltd. |