CN109190043A - Recommended method and device, storage medium, electronic equipment and recommender system - Google Patents
Recommended method and device, storage medium, electronic equipment and recommender system Download PDFInfo
- Publication number
- CN109190043A CN109190043A CN201811046294.6A CN201811046294A CN109190043A CN 109190043 A CN109190043 A CN 109190043A CN 201811046294 A CN201811046294 A CN 201811046294A CN 109190043 A CN109190043 A CN 109190043A
- Authority
- CN
- China
- Prior art keywords
- model
- recall
- recalls
- result
- recalling
- 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 disclosure relates to a kind of recommended method and device, storage medium, electronic equipment and recommender system, to solve the problems, such as that the diversity of recommender system and accuracy are lower in the related technology.The recommended method includes: collection user behavior data, recall in model set multiple are called to recall model according to the user behavior data, it obtains multiple recalling result, wherein, the multiple model of recalling includes leading to recall model and recall model from the master to recall model with different the secondary of rule of recalling；The target that selection meets preset condition in result of recalling that model is recalled from described time recalls result；To the master recall model recall result and the target recalls after result is ranked up and recommends user.
Description
Technical field
This disclosure relates to technical field of data processing, and in particular, to a kind of recommended method and device, storage medium, electricity
Sub- equipment and recommender system.
Background technique
Under Internet scene, recommender system is the indispensable a part of many products, can be not explicit in user
Good personalized ventilation system is provided under conditions of behavior for user.
For example, selling in scene outside, user's high-speed decision is needed, therefore the homepage of application program is it is desirable that user can be given
A variety of demands are provided.It should play the role of traffic distribution by some novelty recommendation results, while need to provide the user with
Accurate personalized recommendation, so as to shorten the commodity selection time of user.This just to recommender system provide higher diversity and
The demand of accuracy.
But in the related technology, recommender system is typically only capable to recall model, example using one kind for a certain specific scene
Such as the proposed algorithm Itembased CF based on article recalls model, cause it is all relatively simple on data granularity and on algorithm,
More accurate and personalized recommendation service can not be provided.
Summary of the invention
Embodiment of the present disclosure main purpose is to provide a kind of recommended method and device, storage medium, electronic equipment and recommendation
System, to solve the problems, such as that the diversity of recommender system and accuracy are lower in the related technology.
To achieve the goals above, embodiment of the present disclosure first aspect provides a kind of recommended method, which comprises
User behavior data is collected, recall in model set multiple is called to recall mould according to the user behavior data
Type obtains multiple recalling result, wherein it is the multiple recall model include it is main recall model and recall model with the master have
Difference recalls the secondary of rule and recalls model；
The target that selection meets preset condition in result of recalling that model is recalled from described time recalls result；
To the master recall model recall result and the target recalls after result is ranked up and recommends user.
Optionally, the target that selection meets preset condition in result of recalling that model is recalled from described time recalls knot
Fruit, comprising:
The daily record data that model is recalled according to described time recalls recalling for model from described time and selects clicking rate in result
And/or conversion ratio meets the result of recalling of threshold condition and recalls result as the target.
It is optionally, described that recall in model set on line multiple is called to recall model according to the user behavior data,
Include:
It is respectively used to call the master to recall model and described according to scheduled flowrate proportioning the user behavior data
It is secondary to recall model, wherein the accounting for recalling model flow into the master is 85% or more, and enters described time and recall model stream
The accounting of amount is greater than 0.
Optionally, the method also includes:
According to the daily record data for time recalling model, determine recall for described time each exposure rate for recalling rule of model and
Conversion ratio；
Exposure rate is higher than first threshold and conversion ratio to recall in model lower than the rule of recalling of second threshold from described time
Deletion.
Optionally, the result of recalling is to recommend the Business Information of user；The master recalls model and uses based on article
Itembased recall rule or the rule of recalling based on user Userbased, recall for described time model using it is following at least
One kind recalling rule: user interest is recalled, correlation rule, matrix decomposition.
Optionally, the method also includes:
Obtain the historical behavior data of user；
It is calculated according to the historical behavior data and recalls that object is similar to recall object with leading；
By each master recall object it is similar recall object storage to the master recall object it is corresponding recall model wait call together
It returns in set.
Optionally, described calculated according to the historical behavior data recalls that object is similar to recall list object with main, wraps
It includes:
It is calculated according to the historical behavior data using the calculation method of following at least one similarity and recalls object with master
It is similar to recall object: cosine similarity, outstanding card similarity, log-likelihood LLR similarity.
Embodiment of the present disclosure second aspect provides a kind of recommendation apparatus, and described device includes:
Data collection module, for collecting user behavior data；
Model calling module, for being called recall in model set multiple to recall mould according to the user behavior data
Type obtains multiple recalling result, wherein it is the multiple recall model include it is main recall model and recall model with the master have
Difference recalls the secondary of rule and recalls model；
As a result screening module, the target that selection meets preset condition in result of recalling for recalling model from described time are called together
Return result；
Sort recommendations module, for the master recall model recall result and the target is recalled result and is ranked up
After recommend user.
Optionally, the result screening module is used for:
The daily record data that model is recalled according to described time recalls recalling for model from described time and selects clicking rate in result
And/or conversion ratio meets the result of recalling of threshold condition and recalls result as the target.
Optionally, the model calling module is used for:
It is respectively used to call the master to recall model and described according to scheduled flowrate proportioning the user behavior data
It is secondary to recall model, wherein the accounting for recalling model flow into the master is 85% or more, and enters described time and recall model stream
The accounting of amount is greater than 0.
Optionally, described device further include:
Index computing module, for according to the daily record data for time recalling model, determining that recall model described time each
Recall the exposure rate and conversion ratio of rule；
Redundant rule elimination module, for by exposure rate be higher than first threshold and conversion ratio lower than second threshold recall rule from
The described deletion recalled in model.
Optionally, the result of recalling is to recommend the Business Information of user；The master recalls model and uses based on article
Itembased recall rule or the rule of recalling based on user Userbased, recall for described time model using it is following at least
One kind recalling rule: user interest is recalled, correlation rule, matrix decomposition.
Optionally, described device further include:
Module is obtained, for obtaining the historical behavior data of user；
Similarity calculation module recalls that object is similar to recall pair with leading for calculating according to the historical behavior data
As；
Memory module, for by each master recall object it is similar recall object and store recall to the master that object is corresponding to call together
Return model wait recall in set.
Optionally, the similarity calculation module is used for:
It is calculated according to the historical behavior data using the calculation method of following at least one similarity and recalls object with master
It is similar to recall object: cosine similarity, outstanding card similarity, log-likelihood LLR similarity.
The embodiment of the present disclosure third aspect provides a kind of non-transitorycomputer readable storage medium, is stored thereon with calculating
Machine program, when which is executed by processor the step of realization first aspect the method.
Embodiment of the present disclosure fourth aspect provides a kind of electronic equipment, comprising:
Memory is stored thereon with computer program；
Processor, for executing the computer program in the memory, to realize first aspect the method
Step.
The 5th aspect of the embodiment of the present disclosure provides a kind of recommender system, and the recommender system includes at least one server,
The log system that the useful daily record data that result is recalled in storage is disposed at least one described server, recalls model set,
And the device for executing first aspect the method.
By adopting the above technical scheme, it at least can achieve following technical effect:
The recommended method that the embodiment of the present disclosure provides is recalled except model using master, is also used and is recalled rule with difference
Then time recall model, improve the diversity of algorithm, and then improve the personalization of recommender system.And mould is recalled for time
Type recall as a result, can will be good, the high result of recalling that scores is merged with the main result of recalling for recalling model, thus
It ensure that the accuracy of recommendation simultaneously on the basis of promoting multifarious.
Other feature and advantage of the embodiment of the present disclosure will the following detailed description will be given in the detailed implementation section.
Detailed description of the invention
Attached drawing is and to constitute part of specification for providing further understanding of the disclosure, with following tool
Body embodiment is used to explain the disclosure together, but does not constitute the limitation to the disclosure.In the accompanying drawings:
Fig. 1 is a kind of flow diagram for recommended method that the embodiment of the present disclosure provides；
Fig. 2 is a kind of structural schematic diagram for recommender system that the embodiment of the present disclosure provides；
Fig. 3 is the schematic diagram for recommending logic on recommender system middle line shown in Fig. 2；
Fig. 4 is that the signal that the target in model recalls result is recalled in feedback system screening time in recommender system shown in Fig. 2
Figure；
Fig. 5 is the structural schematic diagram for another recommender system that the embodiment of the present disclosure provides；
Fig. 6 is a kind of structural schematic diagram for recommendation apparatus that the embodiment of the present disclosure provides；
Fig. 7 is the structural schematic diagram for a kind of electronic equipment that the embodiment of the present disclosure provides.
Specific embodiment
It is described in detail below in conjunction with specific embodiment of the attached drawing to the disclosure.It should be understood that this place is retouched
The specific embodiment stated is only used for describing and explaining the disclosure, is not limited to the disclosure.
A kind of recommended method that the embodiment of the present disclosure provides, as shown in Figure 1, this method comprises:
S101, user behavior data is collected, recall in model set multiple is called to call together according to the user behavior data
Model is returned, obtains multiple recalling result.
Wherein, the multiple model of recalling includes leading to recall model and recall model with the master to recall rule with difference
Time recall model.
It is worth noting that recalling in recommender system for example can be, hot topic is recalled, user interest is recalled, association is advised
Then, collaborative filtering, matrix decomposition and DNN (Deep Neural Network, deep neural network) etc..For different applications
Scene can recall model using different.For example, recommending scene is the major class of application APP homepage or different categories
When page, it is desirable that it allows user to open APP or the information that user wants can be quickly found out when entering major class page, therefore, such scene
Can mainly it be recommended according to the personal preference of user down.In another example when user enters commodity details page, it is desirable that pushed away to user
Other commodity relevant to current commodity are recommended, in this case, can be based on the commodity of current details page, the preference of user is believed
Mode supplemented by breath is recommended.In another example user can enter category column by screening item in category list page or search box
Table page obtains information, if the result that currently screening item or search condition search out is less or comes to nothing, it is desirable that triggering is recommended
Logic carries out information recommendation, in such cases, the extension of current search criteria and user preference information can be combined to be pushed away
It recommends.In the example above, it is possible different that the different ways of recommendation was applicable in recalls model.
Therefore, in the specific implementation, model can be recalled according to actual application scenarios setting master and recalls model with secondary
Recall rule, wherein it is main recall model recall rule compared to time recalling model closer to application scenarios, it is secondary to recall calling together for model
Returning rule can be used as multifarious supplement.In addition, using different models of recalling, being recalled accordingly under different application scene
As a result also different, such as social network sites, this, which recalls result, can be user, for shopping website, this recalls result can be with
It is businessman or commodity.
S102, the target that selection meets preset condition in result of recalling that model is recalled from described time recall result.
It is secondary to recall recalling for model and be mixed into that precision is not high to recall as a result, therefore in result, in order to avoid shadow
Ring the quasi- precision of consequently recommended recommendation results for being shown to user, this step S102 can recall the recalling in result of model from secondary
Filter out good recall as a result, for example time recalling the recalling in result of model from all by preset corresponding condition
Filtering out high conversion rate, perhaps clicking rate is high or recalls that object similarity score is higher to recall result with main.
S103, the master is recalled model recall result and the target recalls after result is ranked up and recommends use
Family.
Specifically, recommending application mostly is to show a recommendation results list to user, belongs to topN recommendation pattern, because
This, after obtaining the main target recalling result He screening for recalling model and recalling result as recommended candidate collection, it is also necessary to
Recommended candidate collection is ranked up, the recommendation results after sequence are finally supplied to upper layer application exhibition by corresponding interface API
Now give user.
Model is recalled using single in the related technology, the preference habit of specific user can only be covered, it can not be useful to institute
The performance that family group can have.And in the technical solution that the embodiment of the present disclosure provides, it recalls except model using master, also makes
Model is recalled with the secondary of rule is recalled with difference, improves the diversity of algorithm, and then improve the individual character of recommender system
Change.And for time recall model recall as a result, can will be good, score and high recall result and recall calling together for model with leading
It returns result to be merged, to ensure that the accuracy of recommendation simultaneously on the basis of promoting multifarious.
Specifically, above-mentioned steps S102 may include: the daily record data that model is recalled according to described time, recall from described time
The result of recalling for selecting clicking rate and/or conversion ratio to meet threshold condition in result of recalling of model recalls knot as the target
Fruit.That is, preset condition described in step S102 refers to the threshold value of the index for measuring the precision for recalling result
Condition, which includes clicking rate and/or conversion ratio, and the clicking rate and conversion ratio can be obtained from daily record data.
Specifically, daily record data includes the exposure log for embodying recommended displaying, embodies and clicks the click checked by user
Log, in the case that recommended object is commodity or businessman, further include embody user whether place an order buy commodity at odd-numbered day will.
To which the clicking rate for recalling object can be calculated based on exposure log and click logs, it is based on click logs and Cheng Dan
The conversion ratio for recalling object can be calculated in log.
Illustratively, Fig. 2 is a kind of schematic diagram for application scenarios that the embodiment of the present disclosure provides, recommendation system as shown in Figure 2
System, comprising: on-line system 11, feedback system 12.Wherein, the on-line system 11 includes recalling model set, as shown in Figure 2
Master recall model, secondary to recall model 1 and recall model N to secondary, the on-line system 11 is used to shunt the behavioral data of user
Model is recalled to each, is recalled as a result, and recording each daily record data for recalling object for recalling result；
The feedback system 12 is used for, and determines that clicking rate and/or conversion ratio meet threshold condition according to the daily record data
Target recall as a result, and the target is recalled result being fused to the master and recall the recalling in result of model.
Based on recommender system shown in Fig. 2, detailed recommended flowsheet is as shown in Figure 3:
After inline system collects the behavioral data of user by upper layer application, behavioral data is diverted to each call together first
Return model, obtain it is each recall model output recall result.Fig. 3 be with based on it is each recall model obtain one recall object into
Row citing obtains recalling object 1 to recalling object N+1.Feedback system recalls mould based on the daily record data in log system, from secondary
The target that screening clicking rate and/or conversion ratio meet threshold condition in result of recalling of type recalls object, and target is recalled object
With it is main recall model recall object as Candidate Set, then Candidate Set input sequencing model is ranked up, will finally be sorted
Recommendation results afterwards are supplied to upper layer application by corresponding interface API and are presented to user.
Optionally, recall in model set on line multiple is called to recall in step S101 according to the user behavior data
Model includes: to be respectively used to call the master to recall model and described according to scheduled flowrate proportioning the user behavior data
It is secondary to recall model, wherein the accounting for recalling model flow into the master is 85% or more, and enters described time and recall model stream
The accounting of amount is greater than 0.Still illustrated with Fig. 2, after collecting user behavior data, can be used 85% or more and 95% with
Under behavioral data flow call it is main recall model, remaining 5% to 15% flow, which enters, time recalls model 1 to N.So that secondary
That recalls model recalls result as the main multifarious supplement for recalling model, rather than as dominating, to ensure that recommendation
Precision.
In a kind of possible implementation of the embodiment of the present disclosure, the method also includes: mould is recalled according to described time
The daily record data of type determines described time and recalls each exposure rate for recalling rule of model and conversion ratio；Exposure rate is higher than first
It is regular from the secondary deletion recalled in model that threshold value and conversion ratio are lower than recalling for second threshold；And/or exposure rate is higher than
Third threshold value and clicking rate are deleted lower than the rule of recalling of the 4th threshold value from recalling in model.That is, if a certain commodity, quilt
Recommend often (i.e. exposure rate is high), but the number checked by user's click is few (i.e. clicking rate is small), then shows to recall the quotient
Recalling for product is regular inaccurate, does not meet the demand of user, therefore, can delete from recalling in model.Similarly, for exposure
Rate is high, and the small commodity of conversion ratio can also be deleted from recalling in model, the accuracy that lift scheme is recommended.It is above-mentioned only to illustrate
Illustrate, corresponding precision threshold can also be set for clicking rate and conversion ratio, judges to recall whether rule needs to delete with this
It removes, i.e., when clicking rate or conversion ratio are less than corresponding precision threshold, recalls rule from recalling to delete in model.
It is still illustrated with Fig. 2, feedback system is calculated each time and recalled model based on the daily record data in log system
The process of index of correlation is as shown in Figure 4.Wherein, each daily record data includes the mark for recalling model, this recalls model
The main mark for recalling object, the mark that the similar object of object is recalled with the master for recalling that model recalls.Wherein, institute in Fig. 4
The id shown is the mark for recalling rule tested on line, identifies for feedback system and recalls rule on the line of daily record data representative
Then, cf_id is the mark for recalling model, and item1 is that this recalls the main item in model, item2 be this recall in model with master
Item similar item, ctr refer to the clicking rate for being exposed to click, and cvr, which refers to, clicks into single conversion ratio.
Specifically, feedback system available main item for each recalling rule in log system recalls model with this and calls together
The relationship of the item returned back out, while the exposure and clicking rate of each object being called back out can be known from log system, this
Sample, can determine according to exposure rate and clicking rate which recall rule be it is accurately applicable, which, which recalls rule, to be deleted.
Illustratively, it is assumed that recalling object in the secondary master recalled in model A is item1, and corresponding similar object of recalling is
Item2, item3, item5, item6, also, by daily record data it is found that item2, item3, item5, item6 are exposed respectively
It 100,200,50,400 times, is clicked respectively 80,100,30,10 times.It can thus be appreciated that the point of item2, item3, item5, item6
Hitting rate is 0.8,0.5,0.6,0.025 respectively.Then in this case, item2 can will be recalled, item3, item5 are called together with master
The object of recalling for returning model is merged and (recommends to show user after being ranked up), and item6 can also will be further recalled
Recall rule from it is secondary recall model A delete.
The above method that the embodiment of the present disclosure provides can be used for immediate distribution field, such as takes out businessman and recommend, also
It is to say, above-mentioned result of recalling is to recommend the Business Information of user.In this case, the master recalls model and uses based on object
The Itembased of product recalls rule or the rule of recalling based on user Userbased, recall for described time model use with down toward
Few one kind recalls rule: user interest is recalled, correlation rule, matrix decomposition.
Optionally, the recommended method that the embodiment of the present disclosure provides further includes that off-line calculation respectively recalls similarity between object
Method, comprising: obtain the historical behavior data of user；It is calculated according to the historical behavior data and recalls that object is similar to call together with leading
Return object；By each master recall object it is similar recall object storage to the master recall object it is corresponding recall model wait recall
In set.
Illustratively, as shown in figure 5, recommender system includes an off-line system 13, major function is used for user's history behavior
The extraction of data.Specifically, the user's history behavioral data that can be used include user in one section of duration (such as 90 days) with
The data of lower behavior: the poi (Point of Interest, point of interest) of user's browsing；User clicks in session session
The data of poi；There are the data for clicking poi behavior after user's search in session.
Wherein, the user's history behavioral data that off-line system extracts can store in the Tool for Data Warehouse based on Hadoop
In Hive table.Wherein, the literary name section of storage is as shown in table 1 below:
Table 1
The user's history behavioral data that off-line system extracts, for calculating the similarity between more new article.For example,
The calculating instrument of similarity can be realized using the distributed computing Spark of MapReduce algorithm, may be implemented to calculate hundred million ranks
The similarity recalled between object item.In the specific implementation, the phase between item can be completed by three MapReduce
It is calculated like degree, each MapReduce completes a calculating operator, and following three calculates operator altogether:
A) preprocessing operator:
B) norm operator:
C) similarity operator: Si,j=similarity (doti,j,ni,nj),
Calculating process is illustrated as example using the outstanding card similarity for specifically calculating vector below, wherein for example need to calculate outstanding person
The vector of card similarity is as follows:
Then it can be calculated by above-mentioned preprocessing operator:
It can be calculated further across norm operator:
Further calculate outstanding card similarity are as follows:
It is above-mentioned only be illustrated with outstanding card similarity, in the specific implementation, can according to calculate cosine similarity,
Outstanding card similarity, any similarity algorithm in log-likelihood LLR similarity calculate the similarity recalled between object.Following table
2 show the corresponding calculating operator of three kinds of common similarity calculating methods.
Table 2
Optionally, passing through similarity calculation, obtaining after leading the similarity list for recalling object and other objects, can incite somebody to action
Similarity list is stored in the following format to key assignments kv system:
key:item
value:item_1:weight_1,item_2:weight_2,……,item_N:weight_N
Wherein, item indicates that master recalls object, and item_1 to item_N is other objects with item similarity, weight
Indicate the similarity between item.Wherein, each master recalls the similarity list of object for being supplied to each of inline system
It recalls model progress article to recall, is improved using key assignments storage and recall efficiency.
Based on identical inventive concept, the embodiment of the present disclosure also provides a kind of recommendation apparatus, for implementing above method reality
The step of recommended method of example offer is provided, as shown in fig. 6, described device includes:
Data collection module 61, for collecting user behavior data；
Model calling module 62, for being called recall in model set multiple to recall mould according to the user behavior data
Type obtains multiple recalling result, wherein it is the multiple recall model include it is main recall model and recall model with the master have
Difference recalls the secondary of rule and recalls model；
As a result screening module 63, for recalling the target of model recalled selection in result and meet preset condition from described time
Recall result；
Sort recommendations module 64, for the master recall model recall result and the target is recalled result and is arranged
User is recommended after sequence.
Using above-mentioned apparatus, which also uses other than using master to recall model and recalls rule with difference
It is secondary to recall model, the diversity of algorithm is improved, and then improve the personalization of recommender system.And model is recalled for time
Recall as a result, can will be good, the high result of recalling that scores is merged with the main result of recalling for recalling model, thus mentioning
It ensure that the accuracy of recommendation on the basis of liter is multifarious simultaneously.
Optionally, the result screening module 63 is used for:
The daily record data that model is recalled according to described time recalls recalling for model from described time and selects clicking rate in result
And/or conversion ratio meets the result of recalling of threshold condition and recalls result as the target.
Optionally, the model calling module 62 is used for:
It is respectively used to call the master to recall model and described according to scheduled flowrate proportioning the user behavior data
It is secondary to recall model, wherein the accounting for recalling model flow into the master is 85% to 95%.
Optionally, described device further include:
Index computing module, for according to the daily record data for time recalling model, determining that recall model described time each
Recall the exposure rate and conversion ratio of rule；
Redundant rule elimination module, for by exposure rate be higher than first threshold and conversion ratio lower than second threshold recall rule from
The described deletion recalled in model.
Optionally, the result of recalling is to recommend the Business Information of user；The master recalls model and uses based on article
Itembased recall rule or the rule of recalling based on user Userbased, recall for described time model using it is following at least
One kind recalling rule: user interest is recalled, correlation rule, matrix decomposition.
Optionally, described device further include:
Module is obtained, for obtaining the historical behavior data of user；
Similarity calculation module recalls that object is similar to recall pair with leading for calculating according to the historical behavior data
As；
Memory module, for by each master recall object it is similar recall object and store recall to the master that object is corresponding to call together
Return model wait recall in set.
Optionally, the similarity calculation module is used for:
It is calculated according to the historical behavior data using the calculation method of following at least one similarity and recalls object with master
It is similar to recall object: cosine similarity, outstanding card similarity, log-likelihood LLR similarity.
About the device in above-described embodiment, wherein modules execute the concrete mode of operation in related this method
Embodiment in be described in detail, no detailed explanation will be given here.
The embodiment of the present disclosure also provides a kind of non-transitorycomputer readable storage medium, is stored thereon with computer journey
The step of sequence, which realizes above-mentioned recommended method when being executed by processor.
The embodiment of the present disclosure also provides a kind of electronic equipment, comprising:
Memory is stored thereon with computer program；
Processor, for executing the computer program in the memory, the step of to realize above-mentioned recommended method.
Illustratively, Fig. 7 is a kind of structural schematic diagram for the above-mentioned electronic equipment that the embodiment of the present disclosure provides.Wherein, the electricity
Sub- equipment may be provided as a kind of server.Referring to Fig. 7, electronic equipment 700 includes processing component 701, further comprises
One or more processors, and the memory resource as representated by memory 702, can be by processing component 701 for storing
The instruction of execution, such as application program.The application program stored in memory 702 may include one or more each
A module for corresponding to one group of instruction.In addition, processing component 701 is configured as executing instruction, to execute above-mentioned recommended method
Step.
Electronic equipment 700 can also include that a power supply module 703 is configured as executing the power supply pipe of electronic equipment 700
Reason, a wired or wireless network interface 704 are configured as electronic equipment being connected to network and an input and output (I/O)
Interface 705.Electronic equipment 700 can be operated based on the operating system for being stored in memory 702, such as Windows
ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or similar.
The embodiment of the present disclosure also provides a kind of recommender system, and the recommender system includes at least one server, it is described extremely
The log system that the useful daily record data that result is recalled in storage is disposed on a few server, recalls model set, Yi Jiyong
In the device for executing above-mentioned recommended method.That is, executing the device of above-mentioned recommended method, model set is recalled in operation, is deposited
The system of storage daily record data can dispose on the same server, can also be deployed on different server respectively, the disclosure pair
This is without limitation.
The preferred embodiment of the disclosure is described in detail in conjunction with attached drawing above, still, the disclosure is not limited to above-mentioned reality
The detail in mode is applied, in the range of the technology design of the disclosure, a variety of letters can be carried out to the technical solution of the disclosure
Monotropic type, these simple variants belong to the protection scope of the disclosure.It is further to note that in above-mentioned specific embodiment
Described in each particular technique feature can be combined in any appropriate way in the case of no contradiction, be
Avoid unnecessary repetition, no further explanation will be given to various combinations of possible ways for the embodiment of the present disclosure.
In addition, any combination can also be carried out between a variety of different embodiments of the disclosure, as long as it is without prejudice to originally
Disclosed thought equally should be considered as disclosure disclosure of that.
Claims (11)
1. a kind of recommended method, which is characterized in that the described method includes:
User behavior data is collected, is called recall in model set multiple to recall model according to the user behavior data, obtained
Result is recalled to multiple, wherein the multiple model of recalling includes that master recalls model and recalls model with different from the master
It recalls the secondary of rule and recalls model；
The target that selection meets preset condition in result of recalling that model is recalled from described time recalls result；
To the master recall model recall result and the target recalls after result is ranked up and recommends user.
2. the method according to claim 1, wherein described time recall recalling in result for model and select from described
The target for meeting preset condition recalls result, comprising:
According to the daily record data for time recalling model, recalled from described time model recall selected in result clicking rate and/or
The result of recalling that conversion ratio meets threshold condition recalls result as the target.
3. the method according to claim 1, wherein described called according to the user behavior data is recalled on line
Multiple in model set recall model, comprising:
The user behavior data according to scheduled flowrate proportioning is respectively used to call the master recalls model and described time is called together
Return model, wherein the accounting for recalling model flow into the master is 85% or more, and enters described time and recall model flow
Accounting is greater than 0.
4. according to the method in any one of claims 1 to 3, which is characterized in that the method also includes:
The daily record data that model is recalled according to described time determines described time and recalls each exposure rate and conversion for recalling rule of model
Rate；
Exposure rate is higher than first threshold and conversion ratio and recalls deleting in model from described time lower than the rule of recalling of second threshold
It removes.
5. according to the method in any one of claims 1 to 3, which is characterized in that the result of recalling is to recommend user
Business Information；The master recalls model and recalls rule using the Itembased based on article or be based on user Userbased
Recall rule, recall model and recall rule using following at least one for described time: user interest is recalled, correlation rule, matrix
It decomposes.
6. according to the method in any one of claims 1 to 3, which is characterized in that the method also includes:
Obtain the historical behavior data of user；
It is calculated according to the historical behavior data and recalls that object is similar to recall object with leading；
By each master recall object it is similar recall object storage to the master recall object it is corresponding recall model wait recall collect
In conjunction.
7. according to the method described in claim 6, it is characterized in that, described calculated according to the historical behavior data is recalled with master
Object is similar to recall list object, comprising:
It is calculated according to the historical behavior data using the calculation method of following at least one similarity and leads that recall object similar
Recall object: cosine similarity, outstanding card similarity, log-likelihood LLR similarity.
8. a kind of recommendation apparatus, which is characterized in that described device includes:
Data collection module, for collecting user behavior data；
Model calling module is obtained for being called recall in model set multiple to recall model according to the user behavior data
Result is recalled to multiple, wherein the multiple model of recalling includes that master recalls model and recalls model with different from the master
It recalls the secondary of rule and recalls model；
As a result screening module, the target that selection meets preset condition in result of recalling for recalling model from described time recall knot
Fruit；
Sort recommendations module, for the master recall model recall result and the target recalls result and is ranked up pusher
It recommends to user.
9. a kind of non-transitorycomputer readable storage medium, is stored thereon with computer program, which is characterized in that the program quilt
The step of processor realizes any one of claims 1 to 7 the method when executing.
10. a kind of electronic equipment characterized by comprising
Memory is stored thereon with computer program；
Processor, for executing the computer program in the memory, to realize any one of claims 1 to 7 institute
The step of stating method.
11. a kind of recommender system, which is characterized in that the recommender system includes at least one server, at least one described clothes
The log system that the useful daily record data that result is recalled in storage is disposed on business device recalls model set, and is used for right of execution
Benefit requires the device of any one of 1 to 7 the method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811046294.6A CN109190043B (en) | 2018-09-07 | 2018-09-07 | Recommendation method and device, storage medium, electronic device and recommendation system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811046294.6A CN109190043B (en) | 2018-09-07 | 2018-09-07 | Recommendation method and device, storage medium, electronic device and recommendation system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109190043A true CN109190043A (en) | 2019-01-11 |
| CN109190043B CN109190043B (en) | 2021-04-23 |
Family
ID=64915397
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811046294.6A Active CN109190043B (en) | 2018-09-07 | 2018-09-07 | Recommendation method and device, storage medium, electronic device and recommendation system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109190043B (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109948023A (en) * | 2019-03-08 | 2019-06-28 | 腾讯科技（深圳）有限公司 | Recommended acquisition methods, device and storage medium |
| CN110275779A (en) * | 2019-06-20 | 2019-09-24 | 北京百度网讯科技有限公司 | A kind of resource acquiring method, device, equipment and storage medium |
| CN110781341A (en) * | 2019-10-09 | 2020-02-11 | 上海麦克风文化传媒有限公司 | Audio album recommendation method and system fusing multi-strategy recall data sets |
| CN110909200A (en) * | 2019-10-09 | 2020-03-24 | 上海麦克风文化传媒有限公司 | Offline and real-time combined audio album recommendation system recall method and system |
| CN110942376A (en) * | 2019-12-02 | 2020-03-31 | 上海麦克风文化传媒有限公司 | Fusion method of real-time multi-recall strategy of audio products |
| CN111161021A (en) * | 2019-12-23 | 2020-05-15 | 叮当快药科技集团有限公司 | Real-time feature-based quick secondary sorting method and tool for recommended commodities |
| CN111178970A (en) * | 2019-12-30 | 2020-05-19 | 微梦创科网络科技（中国）有限公司 | Advertisement delivery method and device, electronic equipment and computer readable storage medium |
| CN111241388A (en) * | 2019-12-13 | 2020-06-05 | 北京三快在线科技有限公司 | Multi-policy recall method and device, electronic equipment and readable storage medium |
| CN111259232A (en) * | 2019-12-03 | 2020-06-09 | 江苏艾佳家居用品有限公司 | Recommendation system optimization method based on personalized recall |
| CN111325598A (en) * | 2020-01-22 | 2020-06-23 | 深圳大学 | Article recommendation method and device and terminal equipment |
| CN111340561A (en) * | 2020-03-04 | 2020-06-26 | 深圳前海微众银行股份有限公司 | Information click rate calculation method, device, equipment and readable storage medium |
| CN111444438A (en) * | 2020-03-24 | 2020-07-24 | 北京百度网讯科技有限公司 | Method, device, equipment and storage medium for determining recall permission rate of recall strategy |
| CN111538901A (en) * | 2020-04-15 | 2020-08-14 | 达而观信息科技（上海）有限公司 | Article recommendation method and device, server and storage medium |
| CN111539210A (en) * | 2020-04-16 | 2020-08-14 | 支付宝(杭州)信息技术有限公司 | Cross-network entity identification method, device, electronic equipment and medium |
| CN111597454A (en) * | 2020-04-02 | 2020-08-28 | 微梦创科网络科技（中国）有限公司 | Account recommendation method and device |
| CN111898028A (en) * | 2020-08-07 | 2020-11-06 | 北京小米移动软件有限公司 | Entity object recommendation method, device and storage medium |
| CN112685633A (en) * | 2020-12-30 | 2021-04-20 | 杭州智聪网络科技有限公司 | Information recommendation method and system based on recall model and prediction model |
| CN112785397A (en) * | 2021-03-09 | 2021-05-11 | 中国工商银行股份有限公司 | Product recommendation method, device and storage medium |
| CN113051413A (en) * | 2019-12-27 | 2021-06-29 | 腾讯科技（北京）有限公司 | Multimedia information processing method and device, electronic equipment and storage medium |
| CN113343089A (en) * | 2021-06-11 | 2021-09-03 | 北京完美赤金科技有限公司 | User recall method, device and equipment |
| CN113407816A (en) * | 2020-03-17 | 2021-09-17 | 北京达佳互联信息技术有限公司 | Object recommendation method and device, electronic equipment and storage medium |
| CN113763010A (en) * | 2020-11-19 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Information pushing method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150149378A1 (en) * | 2013-11-26 | 2015-05-28 | Chegg, Inc. | Job recall services in online education platforms |
| CN106228386A (en) * | 2016-07-12 | 2016-12-14 | 腾讯科技（深圳）有限公司 | A kind of information-pushing method and device |
| CN108198045A (en) * | 2018-01-30 | 2018-06-22 | 东华大学 | The design method of mixing commending system based on e-commerce website data mining |
| US20180181916A1 (en) * | 2013-10-30 | 2018-06-28 | Chegg, Inc. | Correlating jobs with personalized learning activities in online education platforms |
| CN108415928A (en) * | 2018-01-18 | 2018-08-17 | 郝宁宁 | A kind of book recommendation method and system based on weighted blend k- nearest neighbor algorithms |
-
2018
- 2018-09-07 CN CN201811046294.6A patent/CN109190043B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180181916A1 (en) * | 2013-10-30 | 2018-06-28 | Chegg, Inc. | Correlating jobs with personalized learning activities in online education platforms |
| US20150149378A1 (en) * | 2013-11-26 | 2015-05-28 | Chegg, Inc. | Job recall services in online education platforms |
| CN106228386A (en) * | 2016-07-12 | 2016-12-14 | 腾讯科技（深圳）有限公司 | A kind of information-pushing method and device |
| CN108415928A (en) * | 2018-01-18 | 2018-08-17 | 郝宁宁 | A kind of book recommendation method and system based on weighted blend k- nearest neighbor algorithms |
| CN108198045A (en) * | 2018-01-30 | 2018-06-22 | 东华大学 | The design method of mixing commending system based on e-commerce website data mining |
Non-Patent Citations (1)
| Title |
|---|
| 钟良月: ""基于改进的个性化混合推荐算法的研究"", 《中国优秀硕士学位论文全文数据库 社会科学Ⅱ辑》 * |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109948023A (en) * | 2019-03-08 | 2019-06-28 | 腾讯科技（深圳）有限公司 | Recommended acquisition methods, device and storage medium |
| CN109948023B (en) * | 2019-03-08 | 2023-11-03 | 腾讯科技（深圳）有限公司 | Recommended object acquisition method, recommended object acquisition device and storage medium |
| CN110275779A (en) * | 2019-06-20 | 2019-09-24 | 北京百度网讯科技有限公司 | A kind of resource acquiring method, device, equipment and storage medium |
| CN110781341A (en) * | 2019-10-09 | 2020-02-11 | 上海麦克风文化传媒有限公司 | Audio album recommendation method and system fusing multi-strategy recall data sets |
| CN110909200A (en) * | 2019-10-09 | 2020-03-24 | 上海麦克风文化传媒有限公司 | Offline and real-time combined audio album recommendation system recall method and system |
| CN110942376A (en) * | 2019-12-02 | 2020-03-31 | 上海麦克风文化传媒有限公司 | Fusion method of real-time multi-recall strategy of audio products |
| CN110942376B (en) * | 2019-12-02 | 2023-09-01 | 上海麦克风文化传媒有限公司 | Fusion method of real-time multi-recall strategy of audio product |
| CN111259232A (en) * | 2019-12-03 | 2020-06-09 | 江苏艾佳家居用品有限公司 | Recommendation system optimization method based on personalized recall |
| CN111259232B (en) * | 2019-12-03 | 2022-08-12 | 江苏艾佳家居用品有限公司 | Recommendation system optimization method based on personalized recall |
| CN111241388A (en) * | 2019-12-13 | 2020-06-05 | 北京三快在线科技有限公司 | Multi-policy recall method and device, electronic equipment and readable storage medium |
| CN111161021B (en) * | 2019-12-23 | 2021-10-08 | 叮当快药科技集团有限公司 | Quick secondary sorting method for recommended commodities based on real-time characteristics |
| CN111161021A (en) * | 2019-12-23 | 2020-05-15 | 叮当快药科技集团有限公司 | Real-time feature-based quick secondary sorting method and tool for recommended commodities |
| CN113051413A (en) * | 2019-12-27 | 2021-06-29 | 腾讯科技（北京）有限公司 | Multimedia information processing method and device, electronic equipment and storage medium |
| CN111178970A (en) * | 2019-12-30 | 2020-05-19 | 微梦创科网络科技（中国）有限公司 | Advertisement delivery method and device, electronic equipment and computer readable storage medium |
| CN111325598A (en) * | 2020-01-22 | 2020-06-23 | 深圳大学 | Article recommendation method and device and terminal equipment |
| CN111340561A (en) * | 2020-03-04 | 2020-06-26 | 深圳前海微众银行股份有限公司 | Information click rate calculation method, device, equipment and readable storage medium |
| CN113407816A (en) * | 2020-03-17 | 2021-09-17 | 北京达佳互联信息技术有限公司 | Object recommendation method and device, electronic equipment and storage medium |
| CN111444438B (en) * | 2020-03-24 | 2023-09-01 | 北京百度网讯科技有限公司 | Method, device, equipment and storage medium for determining quasi-recall rate of recall strategy |
| CN111444438A (en) * | 2020-03-24 | 2020-07-24 | 北京百度网讯科技有限公司 | Method, device, equipment and storage medium for determining recall permission rate of recall strategy |
| CN111597454A (en) * | 2020-04-02 | 2020-08-28 | 微梦创科网络科技（中国）有限公司 | Account recommendation method and device |
| CN111538901A (en) * | 2020-04-15 | 2020-08-14 | 达而观信息科技（上海）有限公司 | Article recommendation method and device, server and storage medium |
| CN111538901B (en) * | 2020-04-15 | 2023-06-06 | 达而观信息科技（上海）有限公司 | Article recommendation method and device, server and storage medium |
| CN111539210A (en) * | 2020-04-16 | 2020-08-14 | 支付宝(杭州)信息技术有限公司 | Cross-network entity identification method, device, electronic equipment and medium |
| CN111539210B (en) * | 2020-04-16 | 2023-08-11 | 支付宝(杭州)信息技术有限公司 | Cross-network entity identification method and device, electronic equipment and medium |
| CN111898028A (en) * | 2020-08-07 | 2020-11-06 | 北京小米移动软件有限公司 | Entity object recommendation method, device and storage medium |
| CN111898028B (en) * | 2020-08-07 | 2024-04-19 | 北京小米移动软件有限公司 | Entity object recommendation method, device and storage medium |
| CN113763010A (en) * | 2020-11-19 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Information pushing method and device |
| CN112685633A (en) * | 2020-12-30 | 2021-04-20 | 杭州智聪网络科技有限公司 | Information recommendation method and system based on recall model and prediction model |
| CN112785397A (en) * | 2021-03-09 | 2021-05-11 | 中国工商银行股份有限公司 | Product recommendation method, device and storage medium |
| CN112785397B (en) * | 2021-03-09 | 2024-03-29 | 中国工商银行股份有限公司 | Product recommendation method, device and storage medium |
| CN113343089A (en) * | 2021-06-11 | 2021-09-03 | 北京完美赤金科技有限公司 | User recall method, device and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109190043B (en) | 2021-04-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109190043A (en) | Recommended method and device, storage medium, electronic equipment and recommender system | |
| Reddy et al. | Content-based movie recommendation system using genre correlation | |
| US11507551B2 (en) | Analytics based on scalable hierarchical categorization of web content | |
| CN107526807B (en) | Information recommendation method and device | |
| CN104933100B (en) | keyword recommendation method and device | |
| US8751511B2 (en) | Ranking of search results based on microblog data | |
| CN102789462B (en) | A kind of item recommendation method and system | |
| CN108345702A (en) | Entity recommends method and apparatus | |
| CN107066476A (en) | A kind of real-time recommendation method based on article similarity | |
| CN105426514A (en) | Personalized mobile APP recommendation method | |
| CN106600372A (en) | Commodity recommending method and system based on user behaviors | |
| CN105205188A (en) | Method and device for recommending purchase material suppliers | |
| CN102314443B (en) | The modification method of search engine and system | |
| Benouaret et al. | A package recommendation framework for trip planning activities | |
| CN108596695B (en) | Entity pushing method and system | |
| CN102682005A (en) | Method and device for determining preference categories | |
| CN106168980A (en) | Multimedia resource recommends sort method and device | |
| CN106777282B (en) | The sort method and device of relevant search | |
| CN106547864A (en) | A kind of Personalized search based on query expansion | |
| CN108153792A (en) | A kind of data processing method and relevant apparatus | |
| JP4569380B2 (en) | Vector generation method and apparatus, category classification method and apparatus, program, and computer-readable recording medium storing program | |
| CN107316229A (en) | A kind of method and device of online displaying performance project information | |
| CN115712780A (en) | Information pushing method and device based on cloud computing and big data | |
| CN101986301A (en) | Inverse neighbor analysis-based collaborative filtering recommendation system and method | |
| KR101621735B1 (en) | Recommended search word providing method and system |
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 |