CN109086439A - Information recommendation method and device - Google Patents
Information recommendation method and device Download PDFInfo
- Publication number
- CN109086439A CN109086439A CN201810929455.XA CN201810929455A CN109086439A CN 109086439 A CN109086439 A CN 109086439A CN 201810929455 A CN201810929455 A CN 201810929455A CN 109086439 A CN109086439 A CN 109086439A
- Authority
- CN
- China
- Prior art keywords
- information
- candidate
- aggregate
- candidate information
- clicking rate
- 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
The present invention relates to a kind of information recommendation method and device, the information recommendation method includes: to provide information recommendation service and recalling candidate information from information bank;Clicking rate prediction is carried out to the candidate information, pre- sequence information aggregate is generated according to the high candidate information of clicking rate;According to the contextual feature of candidate information in the pre- sequence information aggregate, diversity is carried out to the candidate information in the pre- sequence information aggregate and is reordered, the information aggregate that reorders is obtained;The information recommendation service is provided according to the candidate information in the information aggregate that reorders.Solve the problems, such as that the repetitive rate of information recommendation in the prior art is higher using information recommendation method provided by the present invention and device.
Description
Technical field
The present invention relates to field of computer technology more particularly to a kind of information recommendation methods and device.
Background technique
With the development of internet technology, millions of information can be pushed to user by internet, for example,
When user strolls forum, the topic popular recently to user's recommendation, alternatively, recommending to user and being somebody's turn to do when user reads certain news
Information similar in news content.
Existing information recommendation method is recommended based on clicking rate, i.e., after obtaining candidate information, calculates user
The probability of candidate information, i.e. clicking rate are clicked, candidate information is ranked up according to the clicking rate of candidate information, clicking rate is high
Candidate information preferential recommendation is to user.
Because of situations such as the high candidate information of clicking rate it is possible that repeatability is high, of poor quality, how to reduce letter
The higher problem of repetitive rate present in breath recommendation is still urgently to be resolved.
Summary of the invention
In order to solve the above-mentioned technical problem, it is an object of the present invention to provide a kind of information recommendation method and devices.
Wherein, the technical scheme adopted by the invention is as follows:
In a first aspect, a kind of information recommendation method, comprising: recall candidate letter from information bank to provide information recommendation service
Breath;Clicking rate prediction is carried out to the candidate information, pre- sequence information aggregate is generated according to the high candidate information of clicking rate;According to
The contextual feature of candidate information in the pre- sequence information aggregate, carries out the candidate information in the pre- sequence information aggregate
Diversity reorders, and obtains the information aggregate that reorders;The letter is provided according to the candidate information in the information aggregate that reorders
Cease recommendation service.
Second aspect, a kind of information recommending apparatus, comprising: information recalls module, for for provide information recommendation service from
Candidate information is recalled in information bank;Information pre-ranking module, for carrying out clicking rate prediction to the candidate information, according to click
The high candidate information of rate generates pre- sequence information aggregate;Information reorders module, for according in the pre- sequence information aggregate
The contextual feature of candidate information carries out diversity to the candidate information in the pre- sequence information aggregate and reorders, obtains weight
Sequencing information set;Information recommendation module provides the information for the candidate information in the information aggregate that reorders according to
Recommendation service.
In one exemplary embodiment, the information module that reorders includes: information acquisition unit, for obtaining session page
It has been recommended that information shown in face;Contextual feature extraction unit, for combine get it has been recommended that information, to the walkthrough
Candidate information in sequence information aggregate carries out contextual feature extraction;First clicking rate predicting unit is used for the pre- sequence
The contextual feature of candidate information inputs the first clicking rate prediction model in information aggregate, and prediction obtains the pre- sequencing information collection
The clicking rate of candidate information in conjunction;Ordered sets generation unit is reset, for according to candidate information in the pre- sequence information aggregate
Clicking rate to it is described it is pre- sequence information aggregate in candidate information be ranked up, reorder information aggregate described in generation, described
The information aggregate that reorders includes the slot position of several storage candidate informations, and the candidate information stored in each slot position corresponds to described
Shown in conversation page one it has been recommended that information.
In one exemplary embodiment, the contextual feature extraction unit includes: that set generates subelement, is used for basis
User in the conversation page it has been recommended that the click behavior of information, to getting it has been recommended that information is classified, and/
Or, classifying to the candidate information of slot positions several in the information aggregate that reorders storage, several set are obtained;Basis distribution
Characteristic operation subelement, for dividing for the information in several set and the candidate information in the pre- sequence information aggregate
It Ji Suan not corresponding basic distribution characteristics;Biodiversity Characteristics operation subelement, for according to the basic distribution characteristics being calculated
Carry out Biodiversity Characteristics operation;Contextual feature generates subelement, for being generated by the basic distribution characteristics, Biodiversity Characteristics
The contextual feature of candidate information in the pre- sequence information aggregate.
In one exemplary embodiment, it includes: the first classification subelement that the set, which generates subelement, for according to
User is directed in the conversation page it has been recommended that the click behavior of information, will acquire it has been recommended that information, which is divided to, has clicked letter
Information aggregate is gathered and do not clicked on to breath;And/or second classification subelement, for by the last one be clicked it has been recommended that information draw
Divide to the last one click information set.
In one exemplary embodiment, it includes: that slot position determines subelement that the set, which generates subelement, described in determining
The current slot position to reorder in information aggregate;Third is classified subelement, for by before in the information aggregate that reorders several
The candidate information that slot position is stored is divided to current presentation information aggregate;And/or the 4th classification subelement, for by the rearrangement
The candidate information that previous slot position is stored in sequence information aggregate is divided to previous displaying information aggregate;Wherein, it is preceding several
Slot position is located at before the current slot position in the information aggregate that reorders.
In one exemplary embodiment, the basic distribution characteristics includes: the affiliated level channel distribution characteristics of information, information
Affiliated second level channel distribution characteristics, information labels distribution characteristics, recalls reason distribution characteristics at message subject distribution characteristics.
In one exemplary embodiment, the Biodiversity Characteristics include: diversity factor feature, diversity factor and user's clicking rate group
Close feature, similarity feature, Distribution Entropy feature, cross entropy feature.
In one exemplary embodiment, the rearrangement ordered sets generation unit includes: sorting subunit, for according to
The clicking rate of candidate information carries out the sequence of candidate information in the pre- sequence information aggregate in pre- sequence information aggregate;Storage
Unit will click on the highest candidate information of rate and store to traversal for traversing several slot positions in the information aggregate that reorders
The slot position arrived;Subelement being deleted, being deleted from the pre- sequence information aggregate for will click on the highest candidate information of rate;Notice
First clicking rate predicts subelement.
In one exemplary embodiment, it includes: request reception unit that the information, which recalls module, for receiving client hair
The information recommendation request risen;Request-response unit, for responding information recommendation request from described information storehouse according to recalling
Reason carries out recalling for the candidate information.
In one exemplary embodiment, the information pre-ranking module includes: information characteristics extraction unit, is used for from described
It is extracted in candidate information and obtains information characteristics;Second clicking rate predicting unit, for the information characteristics of the candidate information are defeated
Enter the second clicking rate prediction model, prediction obtains the clicking rate of the candidate information;Pre- ordered set generation unit is used for basis
The clicking rate of the candidate information is ranked up the candidate information, generates the pre- sequence information aggregate.
In one exemplary embodiment, clicking rate prediction model includes the first clicking rate prediction model or the second clicking rate
Prediction model, message sample include it has been recommended that information, input feature vector include it has been recommended that the information characteristics or context of information are special
Sign;Described device further includes model training module, and the model training module includes: sample acquisition unit, is carried for obtaining
The message sample of behavior label, the behavior label are used to indicate the click behavior that user is directed to the message sample;It is defeated
Enter feature extraction unit, for carrying out the extraction of the input feature vector to the message sample;Model training unit is used for basis
Input feature vector and behavior label the guidance designated model of the message sample carry out model training;Model definition unit, being used for will
The designated model of model training is completed as the clicking rate prediction model.
The third aspect, a kind of information recommending apparatus, including processor and memory are stored with computer on the memory
Readable instruction, the computer-readable instruction realize information recommendation method as described above when being executed by the processor.
Fourth aspect, a kind of computer readable storage medium are stored thereon with computer program, the computer program quilt
Processor realizes information recommendation method as described above when executing.
In the above-mentioned technical solutions, two minor sorts will be carried out for candidate information to provide information recommendation service, to solve
The higher problem of repetitive rate present in information recommendation.
Specifically, candidate information is recalled from information bank, this carries out clicking rate prediction to candidate information, according to clicking rate height
Candidate information generate pre- sequence information aggregate, it is right and then according to the contextual feature of candidate information in pre- sequence information aggregate
Candidate information in pre- sequence information aggregate carries out diversity and reorders, and obtains the information aggregate that reorders, and finally by reordering
Candidate information in information aggregate provides information recommendation service, and candidate information has not only carried out clicking rate and sorts in advance as a result, but also
It has carried out diversity to reorder, has guaranteed the diversity for recommending the candidate information of user, efficiently avoid being likely to occur repetition
Property high, ropy candidate information the case where.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
It can the limitation present invention.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and shows and meets implementation of the invention
Example, and in specification together principle for explaining the present invention.
Fig. 1 is a kind of specific implementation schematic diagram of information recommendation method involved in the prior art.
Fig. 2 is the schematic diagram of related implementation environment according to the present invention.
Fig. 3 is a kind of hardware block diagram of server-side shown according to an exemplary embodiment.
Fig. 4 is a kind of flow chart of information recommendation method shown according to an exemplary embodiment.
Fig. 5 be in Fig. 4 corresponding embodiment step 310 in the flow chart of one embodiment.
Fig. 6 be in Fig. 4 corresponding embodiment step 330 in the flow chart of one embodiment.
Fig. 7 be in Fig. 4 corresponding embodiment step 350 in the flow chart of one embodiment.
Fig. 8 is the flow chart of another information recommendation method shown according to an exemplary embodiment.
Fig. 9 be in Fig. 7 corresponding embodiment step 353 in the flow chart of one embodiment.
Figure 10 is that several set involved in Fig. 9 corresponding embodiment, sort in advance information aggregate, the information aggregate that reorders show
It is intended to.
Figure 11 be in Fig. 7 corresponding embodiment step 357 in the flow chart of one embodiment.
Figure 12 is the configuration diagram of information recommendation service in an application scenarios.
Figure 13 is the specific implementation schematic diagram that Figure 12 corresponds to that candidate information in application scenarios repeatedly sorts.
Figure 14 is that Figure 12 corresponds to the generation schematic diagram for recommending news list in application scenarios.
Figure 15 is a kind of block diagram of information recommending apparatus shown according to an exemplary embodiment.
Figure 16 is a kind of hardware block diagram of information recommending apparatus shown according to an exemplary embodiment.
Through the above attached drawings, it has been shown that the specific embodiment of the present invention will be hereinafter described in more detail, these attached drawings
It is not intended to limit the scope of the inventive concept in any manner with verbal description, but is by referring to specific embodiments
Those skilled in the art illustrate idea of the invention.
Specific embodiment
Here will the description is performed on the exemplary embodiment in detail, the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all embodiments consistented with the present invention.On the contrary, they be only with it is such as appended
The example of device and method being described in detail in claims, some aspects of the invention are consistent.
As previously mentioned, existing information recommendation method, carries out according to the probability (i.e. clicking rate) that user clicks candidate information
The recommendation of candidate information is easy to appear situations such as high candidate information repeatability of clicking rate is high, of poor quality.
In order to solve drawbacks described above, as shown in Figure 1, being controlled some diversity are formulated for the high candidate information of clicking rate
Strategy, for example, theme difference between the neighboring candidate information of user is recommended to, in conjunction with the strategy that these are manually formulated, from point
It hits the high candidate information further screening of rate and meets the candidate information of those strategies, and recommend user.
However, not only increasing in above-mentioned recommendation process since the formulation of diversity control strategy depends on artificial realization
The difficulty recommended is added, and has been unfavorable for improving and recommends efficiency.
Therefore, spy of the present invention proposes a kind of information recommendation method, avoids manually formulating diversity control strategy, reduce
The difficulty of recommendation is conducive to improve and recommends efficiency, and can be effectively prevented from the repeatability of information recommendation, correspondingly, this kind
Information recommendation method is suitable for information recommending apparatus, this information recommending apparatus is deployed in the electronics for having von Neumann architecture
In equipment, for example, desktop computer, server etc..
Fig. 2 is a kind of schematic diagram of implementation environment involved in information recommendation method.The implementation environment includes 100 He of terminal
Server-side 200.
Wherein, terminal 100 can be desktop computer, laptop, tablet computer, smart phone or other for visitor
The electronic equipment of family end (such as information recommendation client) operation, herein without limiting.
Terminal 100 and server-side 200 by it is wireless or it is wired pre-establish network connection, to be connected to the network reality by this
Data between existing terminal 100 and server-side 200 are transmitted.For example, the data transmitted include recommended candidate information.
This server-side 200 can be a server, is also possible to the server cluster being made of multiple servers, may be used also
To be the cloud computing center being made of multiple servers.The server is the electronic equipment for providing a user background service,
For example, background service includes Multimedia Recommendation service.
By the interaction of terminal 100 and server-side 200, the client for running on terminal 100 will be initiated to believe to server-side 200
Recommendation request is ceased, and then provides Multimedia Recommendation service by server-side 200, recommended candidate information is pushed into terminal 100
The client of middle operation, to show recommended candidate information to user.
Fig. 3 is a kind of hardware block diagram of server-side shown according to an exemplary embodiment.This server-side is suitable for
Server-side in implementation environment shown by Fig. 2.
It should be noted that this server-side, which is one, adapts to example of the invention, it must not believe that there is provided to this
Any restrictions of the use scope of invention.This server-side can not be construed to need to rely on or must have in Fig. 3 to show
Illustrative server-side 200 in one or more component.
The hardware configuration of this server-side can generate biggish difference due to the difference of configuration or performance, as shown in figure 3, clothes
Business end 200 includes: power supply 210, interface 230, at least a memory 250 and at least central processing unit (CPU, a Central
Processing Units)270。
Wherein, power supply 210 is used to provide operating voltage for each hardware device in server-side 200.
Interface 230 includes an at least wired or wireless network interface 231, at least a string and translation interface 233, at least one defeated
Enter output interface 235 and at least USB interface 237 etc., is used for and external device communication.
The carrier that memory 250 is stored as resource, can be read-only memory, random access memory, disk or CD
Deng the resource stored thereon includes operating system 251, application program 253 and data 255 etc., and storage mode can be of short duration
It stores or permanently stores.Wherein, operating system 251 is for managing and controlling each hardware device in server-side 200 and answer
It can be Windows with program 253 to realize calculating and processing of the central processing unit 270 to mass data 255
ServerTM, Mac OSXTM, UnixTM, LinuxTM, FreeBSDTM etc..Application program 253 be based on operating system 251 it
The upper computer program for completing at least one particular job, may include an at least module (being not shown in Fig. 3), each module
The series of computation machine readable instruction to server-side 200 can be separately included.Data 255 can be stored in disk
Document, audio, video, picture etc..
Central processing unit 270 may include the processor of one or more or more, and be set as through bus and memory
250 communications, for the mass data 255 in operation and processing memory 250.
As described in detail above, memory will be read by central processing unit 270 by being applicable in server-side 200 of the invention
The form of the series of computation machine readable instruction stored in 250 completes information recommendation method.
In addition, also can equally realize the present invention by hardware circuit or hardware circuit combination software, therefore, this hair is realized
The bright combination for being not limited to any specific hardware circuit, software and the two.
Referring to Fig. 4, in one exemplary embodiment, a kind of information recommendation method is suitable for implementation environment shown in Fig. 2
The structure of server-side, the server-side can be as shown in Figure 3.
This kind of information recommendation method can be executed by server-side, may comprise steps of:
Step 310, candidate information is recalled from information bank to provide information recommendation service.
Information recommendation service refers to server-side to client recommended candidate information, in order to which client is that user shows quilt
The candidate information of recommendation.This candidate information can be text information, video information, audio-frequency information, pictorial information etc., this implementation
Example does not make specific limit to the type of candidate information.
It can correspond to different application scenarios accordingly, due to the different type of candidate information, for example, text information can be right
News is answered to read scene, video information can correspond to user's ordering film program scene, and audio-frequency information can correspond to user's program request
Song scene, pictorial information then correspond to user's browsing pictures scene, and therefore, information recommendation method provided in this embodiment can root
It is suitable for different application scenarios according to the different type of candidate information.
As shown in figure 5, in an embodiment in the specific implementation, step 310 may comprise steps of:
Step 311, the information recommendation request that client is initiated is received.
Step 313, response message recommendation request is recalled from information bank according to recalling reason and carry out candidate information.
For client, client provides a request for user and initiates entrance, if user it is expected to carry out information
When recommendation, it can request to initiate herein to trigger relevant operation in entrance, so that client detects this operation, and thus generate letter
Cease recommendation request.
The input module (such as mouse, keyboard, Touch Screen etc.) configured for terminal is different, and entrance is initiated in request to be had
It is distinguished, it is also different to initiate the relevant operation triggered in entrance for request herein.For example, relevant operation includes but is not limited to a little
Hit, move, pull, slide etc..
For example, terminal is smart phone, then initiation entrance is requested to can be the configured Touch Screen of this smart phone
The conversation page of middle presentation illustrates several recommended candidate informations in this conversation page, and user can be by pulling down this session
The page so that client initiates information recommendation request, and then the candidate information that server-side returns is carried out in this conversation page
It updates and shows.Wherein, pulling operation is the relevant operation for requesting to initiate entrance triggering.
For server-side, after client initiates information recommendation request, the request of this information recommendation can be received,
And then information recommendation service is provided for user.
Specifically, candidate information is recalled from information bank first, in accordance with recalling reason, then carry out for candidate information more
Minor sort, recommendation.
Recalling reason includes but is not limited to: just publication, popular, user is interested.That is, for being deposited in information bank
The massive information of storage, the candidate information being called back belongs to rigid release information, alternatively, belong to popular information, or, belong to user
Interested information.
It should be noted that the massive information stored in information bank, is by information publisher's active upload.That is, information
Information to be released is uploaded in the information bank of server-side and stores by publisher, gives this information recommendation in order to server-side and uses
Family.
Step 330, clicking rate prediction is carried out to candidate information, pre- sequencing information is generated according to the high candidate information of clicking rate
Set.
It is appreciated that the candidate information more than one recalled according to reason is recalled, need not all recommend user.This reality
It applies in example, based on the probability that candidate information is clicked by user, i.e. clicking rate, preliminary screening is carried out to the candidate information recalled, it is raw
At pre- sequence information aggregate.
Specifically, as shown in fig. 6, the generating process of sequence information aggregate may comprise steps of in advance:
Step 331, it is extracted from candidate information and obtains information characteristics.
Step 333, the information characteristics of candidate information are inputted into the second clicking rate prediction model, prediction obtains candidate information
Clicking rate.
Step 335, candidate information is ranked up according to the clicking rate of candidate information, generates pre- sequence information aggregate.
Information characteristics are the accurate descriptions to candidate information, to carry out unique identification to candidate information in information.This letter
Breath feature includes but is not limited to the matching degree feature of user characteristics, information content feature and user interest and information.Further
Ground, user characteristics are for describing age of user, gender, occupation, interest etc., and information content feature is for description information theme, letter
Keyword etc. is ceased, the matching degree feature of user interest and information is used to indicate whether information meets user interest, and matching degree is higher,
Then indicate that information more meets user interest.
For example, provide information recommendation service for user A, then the user characteristics of user A are extracted first, then for recalling
Candidate information extracts corresponding information content feature, and the matching degree for calculating recalled candidate information and user A interest is special
Thus the information characteristics that candidate information is directed to user A can be obtained in sign.
From the foregoing, it will be observed that if information recommendation service institute towards user it is different, even if the candidate information recalled is identical, correspondence
Information characteristics also different from, be embodied as different user and the information recommendation service of differentiation be provided, be conducive to promote user's
Recommend experience.
Step 350, according to the contextual feature of candidate information in pre- sequence information aggregate, in pre- sequence information aggregate
Candidate information carries out diversity and reorders, and obtains the information aggregate that reorders.
It is appreciated that recommend user when candidate information, that is, it is considered as it has been recommended that information, in order to avoid existing in information recommendation
Repeated problem, then need to consider it has been recommended that influence of the information to candidate information, for example, the theme of candidate information is different from
The theme etc. of recommendation information.
Contextual feature, for describe candidate information with it has been recommended that correlation between information, and description candidate information
Between correlation.
Diversity reorders as a result, not only for it has been recommended that information, it is contemplated that candidate information with it has been recommended that information multiplicity
Property, and between candidate information, it is contemplated that the diversity between candidate information.
Step 370, information recommendation service is provided according to the candidate information in the information aggregate that reorders.
For client, after the candidate information that server-side has pushed in the information aggregate that reorders, it can receive
To the candidate information in the information aggregate that reorders, and then recommend user.
For example, in the conversation page that client is presented, in the information aggregate that will be reordered in the form of session list
Several candidate informations show user.
By process as described above, multifarious candidate information is provided for user, realizes the diversity of information recommendation,
It is effectively prevented from the case where being likely to occur repeated height, ropy candidate information.
Referring to Fig. 7, in one exemplary embodiment, step 350 may comprise steps of:
Step 351, it obtains shown in conversation page it has been recommended that information.
Conversation page, the candidate information in the information aggregate that reorders for showing server-side push.
For client, as client runs on terminal, the candidate information that conversation page just pushes server-side
It is correspondingly presented in the screen that terminal is configured, so that candidate information is recommended to user.
And for server-side, with the displaying of conversation page, the candidate information in the information aggregate that reorders can be considered
It has been recommended that information shown in conversation page.
Step 353, in conjunction with getting it has been recommended that information, carries out context to the candidate information in pre- sequence information aggregate
Feature extraction.
Step 355, the contextual feature of candidate information in pre- sequence information aggregate is inputted into the first clicking rate prediction model,
Prediction obtains the clicking rate of candidate information in pre- sequence information aggregate.
Step 357, according to the clicking rate of candidate information in pre- sequence information aggregate to the candidate in pre- sequence information aggregate
Information is ranked up, and generates the information aggregate that reorders.
Wherein, the information aggregate that reorders includes the slot position of several storage candidate informations, the candidate stored in each slot position
Information is corresponding to one shown in conversation page it has been recommended that information.
It is appreciated that client will constantly initiate information recommendation request, and reorder letter as a result, during information recommendation
The candidate information that several slot positions are stored in breath set will be continuously updated, so that it has been recommended that letter shown in conversation page
Breath correspondingly constantly variation therewith.
Under the action of above-described embodiment, the diversity of candidate information in the information aggregate that reorders is realized, is pushed away for information
The diversity recommended provides necessary foundation.
For the first clicking rate prediction model and the second clicking rate prediction model, be for realizing
Clicking rate prediction, difference are only that input object difference, and output object is also different.
Here, in order to better describe the first clicking rate prediction model and the second clicking rate prediction model in model training mistake
General character in journey is such as given a definition for above-mentioned difference and is illustrated.
Wherein, clicking rate prediction model includes the first clicking rate prediction model or the second clicking rate prediction model.
Message sample includes it has been recommended that information.
Correspondingly, input feature vector includes information characteristics or contextual feature it has been recommended that information.
Correspondingly, as shown in figure 8, in one exemplary embodiment, method as described above can also include that clicking rate is pre-
The model training process of model is surveyed, the model training process of this clicking rate prediction model may comprise steps of:
Step 410, the message sample of carrying behavior label is obtained.
Wherein, behavior label is used to indicate the click behavior that user is directed to message sample.
That is, for show in conversation page it has been recommended that information, if user clicks it has been recommended that information,
Behavior label indicates it has been recommended that information is clicked by user., whereas if user does not click on it has been recommended that information, then behavior label refers to
Show it has been recommended that information is not clicked by user.
Step 430, the extraction of input feature vector is carried out to message sample.
Step 450, model training is carried out according to the input feature vector of message sample and behavior label guidance designated model.
Wherein, designated model includes but is not limited to: the machines such as logistic regression, support vector machines, random forest, neural network
Learning model.
Model training, substantially according to the input feature vector of message sample and behavior label to the model parameter of designated model
It optimizes, obtains making the convergent optimal model parameters of designated model with study.
Specifically, the model parameter of random initializtion designated model will work as the input feature vector and row of previous message sample
Designated model is inputted for label, it is assumed that the model parameter of random initializtion fails to restrain designated model, then to random initializtion
Model parameter be updated, and the input feature vector of latter information sample and behavior label are input to designated model.
Such iteration, until the number of iterations reaches iteration threshold or the model parameter of update restrains designated model,
Complete the model training of designated model.
Wherein, iteration threshold can neatly be adjusted according to the actual needs of application scenarios.For example, being wanted to forecasting accuracy
It asks in higher application scenarios, biggish iteration threshold is set, alternatively, in the application scenarios more demanding to predetermined speed, if
Set lesser iteration threshold.
Step 470, the designated model of model training will be completed as clicking rate prediction model.
It is finished to model training, designated model converges to clicking rate prediction model, and optimal model parameters are used as a little
The input parameter of rate prediction model is hit, to predict to obtain the clicking rate of candidate information based on this clicking rate prediction model.
It is described herein to be, clicking rate prediction, substantially according to the information characteristics of candidate information or contextual feature,
Clicking rate prediction model is called to predict to obtain the probability of the affiliated behavior label of this candidate information.
For example, it is assumed that behavior label is 0, instruction user will not click candidate information, and behavior label is 1, indicate user's meeting
Click candidate information.
Further, if the probability that candidate information belongs to behavior label 0 is P0, the probability for belonging to behavior label 1 is P1,
If P0>P1, prediction user will not click this candidate information, then clicking rate is 0, whereas if P0<P1, prediction user can point
This candidate information is hit, then using P1 as the clicking rate of candidate information.
Under the cooperation of above-described embodiment, the model training based on designated model is realized, i.e., by message sample
Machine learning enables clicking rate prediction model to predict that user is directed to the click behavior of candidate information well, and then effectively
The accuracy of ground raising clicking rate prediction model.
In addition, being based on machine learning, avoids manually formulating diversity control strategy, reduce recommendation difficulty, be conducive to take
Business end provides thousand people, thousand face, more accurate, more personalized diversity recommendation service.
Referring to Fig. 9, in one exemplary embodiment, step 353 may comprise steps of:
Step 3531, according to user in conversation page it has been recommended that the click behavior of information, to getting it has been recommended that
Information is classified, and/or, the candidate information of several slot position storages is classified in reorder information aggregate, is obtained several
Set.
As shown in Figure 10, conversation page 501, show fixed number it has been recommended that information.For example, fixed number is 5.
Pre- sequence information aggregate 502, includes several candidate informations.
Reorder information aggregate 503, is used to store the slot position of candidate information comprising fixed number.Each slot position corresponds to
One in conversation page 501 it has been recommended that information, for example, fixed number is 5.
It is several set 504, including click information set 5041, do not click on information aggregate 5042, current presentation information collection
Close 5043, the last one click information set 5044, previous displaying information aggregate 5045.
Using candidate information by news for example, for conversation page 501 show it has been recommended that information, if it has been recommended that
Information is clicked by user, then will be it has been recommended that information is divided to click information set 5041.If it has been recommended that information is not by user
It clicks, then it will be it has been recommended that information, which is divided to, click on information aggregate 5042.By the last one be clicked it has been recommended that information divide
To the last one click information set 5044.
By above three set, reflects it has been recommended that the click of information is distributed, the interest preference of user is indicated with this.
In order to conversation page 501 can show fixed number it has been recommended that information, fixed in the information aggregate 503 that reorders
Several slot positions needs to store candidate information, and the candidate information stored in each slot position is both from pre- sequence information aggregate 502
Candidate information.
Assuming that current slot position 5031 not yet stores candidate information, and it is located at preceding 3 slot positions before current slot position 5031
5032,5033,5034 candidate information has been stored, then the candidate information stored preceding 3 slot positions 5032,5033,5034 divides
To current presentation information aggregate 5043.The candidate information that preceding 1 slot position 5032 is stored is divided to previous displaying information collection
Close 5045.
By above-mentioned two set, reflects the correlation being recommended between the candidate information of same user, avoid letter
The high candidate information of duplicating property in breath recommendation.
Step 3533, it for the information in several set and the candidate information in pre- sequence information aggregate, calculates separately
Corresponding basis distribution characteristics.
As previously mentioned, several set, including click information set, do not click on information aggregate, current presentation information aggregate,
The last one click information set, previous displaying information aggregate etc..Correspondingly, information can be it has been recommended that believing in several set
Breath is also possible to reset the candidate information that some slot position of ordered sets is stored.
Wherein, basic distribution characteristics includes: the affiliated level channel distribution characteristics of information, the affiliated second level channel distribution spy of information
Sign, recalls reason distribution characteristics at message subject distribution characteristics, information labels distribution characteristics.
For example, level channel includes " political situation of the time ", " military affairs ", " science and technology ", " sport ", " amusement ", " education ", " trip
Trip ", " cuisines ", " health " etc..
Second level channel is the subdivision to level channel, for example, level channel " sport " can be subdivided into " football ", " basket
The second levels channel such as ball ", " swimming ", " diving ".
Message subject is equivalent to the classification of information, for example, massive information is directed to, by cluster, by similar massive information
500 classification are divided into, this 500 classification can be considered the message subject of massive information.
Information labels can refer to the keyword of information, may also mean that the publisher of information, can also refer to information institute
The mood of expression.For example, mood expressed by song is sadness when information is song, then the information of sad visual song thus
Label, or, singer, the creation of words and music person of song can be used as the information labels of this song.
Reason is recalled, popular, just publication, user are interested etc. as previously mentioned, can be.
Certainly, according to the actual needs of application scenarios, basic distribution characteristics can also include information word distribution characteristics, information
Content distribution feature etc., the present embodiment not constitute this and limit.
Specifically, it for several set and pre- sequence information aggregate, is calculated as follows respectively.
(1) the affiliated level channel distribution characteristics of information:Wherein, ncFor level-one frequency
Road sum,
(2) the affiliated second level channel distribution characteristics of information:Wherein, nsFor subdivision frequency
Road sum,
(3) message subject distribution characteristics:Wherein, ntIt is total for message subject,
(4) information labels distribution characteristics:Wherein, ngIt is total for information labels,
(5) reason distribution characteristics is recalled:Wherein, nrTo recall reason sum,
Step 3535, Biodiversity Characteristics operation is carried out according to the basic distribution characteristics being calculated.
Wherein, Biodiversity Characteristics include: diversity factor feature, diversity factor and user's clicking rate assemblage characteristic, similarity feature,
Distribution Entropy feature, cross entropy feature.
Specifically, it for several set and the pre- information aggregate basic distribution characteristics accordingly that sorts, is counted as follows respectively
It calculates.
For ease of description, in several set, click information set referred to as gathers 1, does not click on information aggregate referred to as
For set 2, current presentation information aggregate referred to as gathers 3, the last one click information set referred to as gathers 4, previous exhibition
Show that information aggregate referred to as gathers 5.
(1) diversity factor feature:
Calculate separately set 1-5 and the affiliated level channel distribution characteristics of the pre- sequence corresponding information of information aggregate, information institute
Belong to second level channel distribution characteristics, message subject distribution characteristics, information labels distribution characteristics, recall dividing between reason distribution characteristics
Cloth is poor, obtains diversity factor feature, is denoted as Diff respectivelyC(j), DiffS(j), DiffT(j), DiffG(j), DiffR(j), 1≤j≤
5。
It is to gather the distribution difference between 1 and the pre- sequence affiliated level channel distribution characteristics of the corresponding information of information aggregate
Example, it is assumed that the affiliated level channel distribution characteristics of information of sequence information aggregate is in advance Collection
Close 1 the affiliated level channel distribution characteristics of information be Then sort information aggregate and collection in advance
The distribution closed between the 1 affiliated level channel distribution characteristics of information is poor, i.e. diversity factor feature are as follows:
(2) diversity factor and user's clicking rate assemblage characteristic:
The diversity factor feature obtained according to (1) and user are for information in set 1-5 in affiliated level channel, affiliated
Second level channel, message subject, information labels and the clicking rate recalled in reason are distributed, and are calculated diversity factor and are combined with user's clicking rate
Feature is denoted as DiffCTR respectivelyC(j), DiffCTRS(j)DiffCTRT(j), DiffCTRG(j), DiffCTRR(j), 1≤j≤
5。
With diversity factor feature Diffc(1) it is distributed as information in set 1 in the clicking rate of affiliated level channel with user
Example, it is assumed that the distribution between pre- sequence information aggregate and the affiliated level channel distribution characteristics of information for gathering 1 is poor, i.e., diversity factor is special
SignClicking rate of the user for information in set 1 in affiliated level channel is divided
Cloth isWherein, ncIt is total for level channel,
Correspondingly, diversity factor and user's clicking rate assemblage characteristic are as follows:
(3) similarity feature:
Calculate separately the affiliated level channel distribution characteristics of pre- sequence information aggregate information corresponding with set 1-5, affiliated two
It grade channel distribution characteristics, message subject distribution characteristics, information labels distribution characteristics and recalls similar between reason distribution characteristics
Degree, obtains similarity feature, is denoted as Simi respectivelyC(j), SimiS(j), SimiT(j), SimiG(j), SimiR(j), 1≤j≤5.
It is false for gathering the similarity between 1 and the affiliated level channel distribution of the pre- sequence corresponding information of information aggregate
If the affiliated level channel distribution characteristics of information of sequence information aggregate is in advance Set 1
The affiliated level channel distribution characteristics of information isThen similarity feature between the two are as follows:
SimiC(1)=cos (Cd, Ch)。
Wherein, cos () is the vector cosine similarity function of standard.
(4) Distribution Entropy feature:
It is special to calculate separately the affiliated second level channel distribution of the affiliated level channel distribution characteristics of the corresponding information of set 1-5, information
Sign, message subject distribution characteristics, information labels distribution characteristics and the Distribution Entropy for recalling reason distribution characteristics obtain Distribution Entropy spy
Sign, is denoted as Etp respectivelyC(j), EtpS(j), EtpT(j), EtpG(j), EtpR(j), 1≤j≤5.
For gathering 1 Distribution Entropy of the affiliated level channel distribution characteristics of information, it is assumed that gather the 1 affiliated level-one of information
Channel distribution characteristics isThen corresponding Distribution Entropy feature are as follows:
Wherein, log () is logarithmic function.
(5) cross entropy feature:
Calculate separately the affiliated level channel distribution characteristics of pre- sequence information aggregate information corresponding with set 1-5, information institute
Belong to second level channel distribution characteristics, message subject distribution characteristics, information labels distribution characteristics and recalls between reason distribution characteristics
Cross entropy obtains cross entropy feature, is denoted as Ctp respectivelyC(j), CtpS(j), CtpT(j), CtpG(j), CtpR(j), 1≤j≤5.
For the cross entropy of the affiliated level channel distribution of information to gather 1, it is assumed that the information institute of pre- sequence information aggregate
Belonging to level channel distribution characteristics isThe affiliated level channel distribution characteristics of information of set 1
ForThen corresponding cross entropy feature are as follows:
Wherein, log () is logarithmic function.
Step 3537, by basic distribution characteristics, Biodiversity Characteristics generate candidate information in pre- sequence information aggregate up and down
Literary feature.
The contextual feature of candidate information is to combine above-mentioned each basic distribution characteristics, each Biodiversity Characteristics.
By the above process, the contextual feature for realizing candidate information is extracted, and features recommended time from multiple dimensions
The diversity between information is selected, the diversity in the information aggregate that reorders between candidate information has both been considered, it is also contemplated that pushed away
The diversity between information and candidate information is recommended, repeating for candidate information can be effectively avoided, and can be effectively
The interest diversity preference of user itself is portrayed, the diversity recommendation service of differentiation is provided for different user, is conducive to be promoted
The recommendation of user is experienced.
Figure 11 is please referred to, in one exemplary embodiment, step 357 may comprise steps of:
Step 3571, it is carried out according to the clicking rate of candidate information in pre- sequence information aggregate candidate in pre- sequence information aggregate
The sequence of information.
Step 3573, traversal reorders several slot positions in information aggregate, will click on the highest candidate information of rate store to
The slot position traversed.
Step 3575, it will click on the highest candidate information of rate to delete from pre- sequence information aggregate.
As previously mentioned, the information aggregate that reorders includes the slot position of several storage candidate informations, each slot position corresponds to meeting
Talk about the page shown in one it has been recommended that information.
So, reorder the generating process of information aggregate, substantially deposits the candidate information in pre- sequence information aggregate
Store up several slot positions into the information aggregate that reorders.
As shown in Figure 10, for the candidate information in the information aggregate 502 that sorts in advance, the highest candidate information of clicking rate
First slot position 5034 into the information aggregate 503 that reorders will preferentially be stored.
Correlation it is appreciated that consideration is reordered in information aggregate 503 between candidate information, the high candidate of clicking rate time
Information may not can store second slot position 5033 into the information aggregate 503 that reorders, for example, the candidate high when clicking rate time
Information and clicking rate highest candidate information, then can not quilts due to there is the phenomenon that repeating when the theme of the two is consistent
Recommend same user.
For this purpose, will click on the highest candidate information of rate after deleting in pre- sequence information aggregate, then execution step is jumped
353, the contextual feature of candidate information in pre- sequence information aggregate is extracted again.
And so on, until the last one slot position 5035 in the information aggregate 503 that reorders stores candidate information.
The candidate information in the information aggregate 503 that reorders can be pushed to client as a result, to be presented in client
Conversation page 501 in show, and then complete information recommendation service.
Figure 12~Figure 14 is a kind of specific implementation schematic diagram of information recommendation method in an application scenarios.The application scenarios
In, terminal is smart phone, for news reader operation.
As news reader is in the operation of smart phone, news recommendation request will be initiated to server-side, so that server-side
News Recommendation Service Based is provided for user.
For server-side, provided News Recommendation Service Based includes three modules: contextual feature abstraction module
601, off-line model training module 602 and online news recommending module 603, as shown in figure 12.
Wherein, contextual feature abstraction module 601 is according to the candidate news 605 and history recalled in real time it has been recommended that news
604, extract the contextual feature of candidate information.
Message sample training clicking rate prediction model 606 of the off-line model training module 602 based on magnanimity, is realized pair with this
The clicking rate of candidate news sorts in advance and diversity reorders.
Specifically, as shown in figure 13, it by executing step 701 to step 702, is chosen from the candidate information recalled in real time
Clicking rate higher multiple candidate informations form pre- sequence information aggregate.For example, 100 candidate informations.
By executing step 703 to step 707, in conjunction with the contextual feature of candidate information, from pre- sequence information aggregate
The higher multiple candidate informations of clicking rate are chosen, the information aggregate that reorders is formed.
Online news recommending module 603 forms then based on the candidate information in the information aggregate that reorders and recommends news list
607, as shown in figure 14, and news reader is finally pushed to, so that this is included pushing away for multiple candidate informations by news reader
It recommends news list and shows user.
In this application scene, by the diversified candidate news of offer, the reading experience of user can be improved well,
It improves user to be directed to the clicking rate of candidate news and read duration, is conducive to create preferable product economy benefit.
Following is apparatus of the present invention embodiment, can be used for executing information recommendation method according to the present invention.For this
Undisclosed details in invention device embodiment, please refers to the embodiment of the method for information recommendation method according to the present invention.
Figure 15 is please referred to, in one exemplary embodiment, a kind of information recommending apparatus 900 includes but is not limited to: information is called together
Return module 910, information pre-ranking module 930, information reorder module 950 and information recommendation module 970.
Wherein, information recalls module 910 for provide information recommendation service and recall candidate information from information bank.
Information pre-ranking module 930 is used to carry out clicking rate prediction to candidate information, according to the high candidate information of clicking rate
Generate pre- sequence information aggregate.
Information reorder module 950 for according in advance sort information aggregate in candidate information contextual feature, to walkthrough
Candidate information in sequence information aggregate carries out diversity and reorders, and obtains the information aggregate that reorders.
Information recommendation module 970 is used to provide information recommendation service according to the candidate information in the information aggregate that reorders.
It should be noted that information recommending apparatus provided by above-described embodiment is when carrying out information recommendation processing, only with
The division progress of above-mentioned each functional module can according to need and for example, in practical application by above-mentioned function distribution by not
Same functional module is completed, i.e., the internal structure of information recommending apparatus will be divided into different functional modules, to complete above retouch
The all or part of function of stating.
In addition, the embodiment of information recommending apparatus and information recommendation method provided by above-described embodiment belongs to same structure
Think, the concrete mode that wherein modules execute operation is described in detail in embodiment of the method, no longer superfluous herein
It states.
Figure 16 is please referred to, in one exemplary embodiment, a kind of information recommending apparatus 1000, including an at least processor
1001, an at least memory 1002 and at least a communication bus 1003.
Wherein, computer-readable instruction is stored on memory 1002, processor 1001 is read by communication bus 1003
The computer-readable instruction stored in memory 1002.
The information recommendation method in the various embodiments described above is realized when the computer-readable instruction is executed by processor.
In one exemplary embodiment, a kind of computer readable storage medium, is stored thereon with computer program, the calculating
The information recommendation method in the various embodiments described above is realized when machine program is executed by processor.
Above content, preferable examples embodiment only of the invention, is not intended to limit embodiment of the present invention, this
Field those of ordinary skill central scope according to the present invention and spirit can be carried out very easily corresponding flexible or repaired
Change, therefore protection scope of the present invention should be subject to protection scope required by claims.
Claims (13)
1. a kind of information recommendation method characterized by comprising
Candidate information is recalled from information bank to provide information recommendation service;
Clicking rate prediction is carried out to the candidate information, pre- sequence information aggregate is generated according to the high candidate information of clicking rate;
According to the contextual feature of candidate information in the pre- sequence information aggregate, to the candidate in the pre- sequence information aggregate
Information carries out diversity and reorders, and obtains the information aggregate that reorders;
The information recommendation service is provided according to the candidate information in the information aggregate that reorders.
2. the method as described in claim 1, which is characterized in that described according to candidate information in the pre- sequence information aggregate
Contextual feature carries out diversity to the candidate information in the pre- sequence information aggregate and reorders, obtains the information collection that reorders
It closes, comprising:
It obtains shown in conversation page it has been recommended that information;
In conjunction with getting it has been recommended that information, carries out contextual feature to the candidate information in the pre- sequence information aggregate and mention
It takes;
The contextual feature of candidate information in the pre- sequence information aggregate is inputted into the first clicking rate prediction model, prediction obtains
The clicking rate of candidate information in the pre- sequence information aggregate;
According to the clicking rate of candidate information in the pre- sequence information aggregate to the candidate information in the pre- sequence information aggregate
It being ranked up, reorder information aggregate described in generation, and the information aggregate that reorders includes the slot position of several storage candidate informations,
The candidate information stored in each slot position is corresponding to one shown in the conversation page it has been recommended that information.
3. method according to claim 2, which is characterized in that it is that the combination is got it has been recommended that information, to the walkthrough
Candidate information in sequence information aggregate carries out contextual feature extraction, comprising:
According to user in the conversation page it has been recommended that the click behavior of information, to getting it has been recommended that information is divided
Class, and/or, classify to the candidate information of slot positions several in the information aggregate that reorders storage, obtains several set;
For the information in several set and the candidate information in the pre- sequence information aggregate, corresponding basis is calculated separately
Distribution characteristics;
Biodiversity Characteristics operation is carried out according to the basic distribution characteristics being calculated;
The context for generating candidate information in the pre- sequence information aggregate by the basic distribution characteristics, Biodiversity Characteristics is special
Sign.
4. method as claimed in claim 3, which is characterized in that described to be directed in the conversation page according to user it has been recommended that letter
The click behavior of breath obtains several set to getting it has been recommended that information is classified, comprising:
It is directed in the conversation page according to the user it has been recommended that the click behavior of information, will acquire it has been recommended that information is drawn
Divide to click information set and does not click on information aggregate;And/or
By the last one be clicked it has been recommended that information is divided to the last one click information set.
5. method as claimed in claim 3, which is characterized in that described to be stored to slot positions several in the information aggregate that reorders
Candidate information classify, obtain several set, comprising:
The current slot position to be reordered in information aggregate described in determination;
The candidate information that several slot positions preceding in the information aggregate that reorders are stored is divided to current presentation information aggregate;
And/or
The candidate information that previous slot position is stored in the information aggregate that reorders is divided to previous displaying information aggregate;
Wherein, several preceding slot positions are located at before the current slot position in the information aggregate that reorders.
6. method as claimed in claim 3, which is characterized in that the basis distribution characteristics includes: the affiliated level channel of information
Distribution characteristics, information affiliated second level channel distribution characteristics, information labels distribution characteristics, recall reason at message subject distribution characteristics
Distribution characteristics.
7. method as claimed in claim 3, which is characterized in that the Biodiversity Characteristics include: diversity factor feature, diversity factor with
User's clicking rate assemblage characteristic, similarity feature, Distribution Entropy feature, cross entropy feature.
8. method according to claim 2, which is characterized in that described according to candidate information in the pre- sequence information aggregate
Clicking rate is ranked up the candidate information in the pre- sequence information aggregate, and reorder information aggregate described in generation, comprising:
Candidate information in the pre- sequence information aggregate is carried out according to the clicking rate of candidate information in the pre- sequence information aggregate
Sequence;
Several slot positions to be reordered in information aggregate described in traversal, will click on the highest candidate information of rate and store to the slot traversed
Position;
It will click on the highest candidate information of rate to delete from the pre- sequence information aggregate;
It jumps and executes that the combination is got it has been recommended that information, carries out the candidate information in the pre- sequence information aggregate
The step of following traits extract.
9. method as claimed in any one of claims 1 to 8, which is characterized in that described is to provide information recommendation service from information
Candidate information is recalled in library, comprising:
Receive the information recommendation request that client is initiated;
It responds information recommendation request and carries out recalling for the candidate information according to recalling reason from described information storehouse.
10. method as claimed in any one of claims 1 to 8, which is characterized in that described to carry out clicking rate to the candidate information
Prediction generates pre- sequence information aggregate according to the high candidate information of clicking rate, comprising:
It is extracted from the candidate information and obtains information characteristics;
The information characteristics of the candidate information are inputted into the second clicking rate prediction model, prediction obtains the click of the candidate information
Rate;
The candidate information is ranked up according to the clicking rate of the candidate information, generates the pre- sequence information aggregate.
11. method as claimed in claim 10, which is characterized in that clicking rate prediction model includes the first clicking rate prediction model
Or the second clicking rate prediction model, message sample include it has been recommended that information, input feature vector include it has been recommended that the information of information is special
Sign or contextual feature;
The method also includes:
The message sample of carrying behavior label is obtained, the behavior label is used to indicate user for the message sample
Click behavior;
The extraction of the input feature vector is carried out to the message sample;
Model training is carried out according to the input feature vector of the message sample and behavior label guidance designated model;
The designated model of model training will be completed as the clicking rate prediction model.
12. a kind of information recommending apparatus characterized by comprising
Information recalls module, for recalling candidate information from information bank to provide information recommendation service;
Information pre-ranking module, it is raw according to the high candidate information of clicking rate for carrying out clicking rate prediction to the candidate information
At pre- sequence information aggregate;
Information reorders module, for the contextual feature according to candidate information in the pre- sequence information aggregate, to described pre-
Candidate information in sequencing information set carries out diversity and reorders, and obtains the information aggregate that reorders;
Information recommendation module provides the information recommendation service for the candidate information in the information aggregate that reorders according to.
13. a kind of information recommending apparatus characterized by comprising
Processor;And
Memory is stored with computer-readable instruction on the memory, and the computer-readable instruction is held by the processor
The information recommendation method as described in any one of claims 1 to 11 is realized when row.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810929455.XA CN109086439B (en) | 2018-08-15 | 2018-08-15 | Information recommendation method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810929455.XA CN109086439B (en) | 2018-08-15 | 2018-08-15 | Information recommendation method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109086439A true CN109086439A (en) | 2018-12-25 |
CN109086439B CN109086439B (en) | 2022-02-25 |
Family
ID=64793572
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810929455.XA Active CN109086439B (en) | 2018-08-15 | 2018-08-15 | Information recommendation method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109086439B (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109981785A (en) * | 2019-04-02 | 2019-07-05 | 北京字节跳动网络技术有限公司 | Method and apparatus for pushed information |
CN110110233A (en) * | 2019-05-09 | 2019-08-09 | 网易传媒科技(北京)有限公司 | Information processing method, device, medium and calculating equipment |
CN110659388A (en) * | 2019-10-10 | 2020-01-07 | 北京奇艺世纪科技有限公司 | To-be-recommended information screening method and device, electronic equipment and storage medium |
CN110825974A (en) * | 2019-11-22 | 2020-02-21 | 厦门美柚股份有限公司 | Recommendation system content ordering method and device |
CN111008278A (en) * | 2019-11-22 | 2020-04-14 | 厦门美柚股份有限公司 | Content recommendation method and device |
CN111381909A (en) * | 2018-12-27 | 2020-07-07 | 北京字节跳动网络技术有限公司 | Page display method and device, terminal equipment and storage medium |
CN111382346A (en) * | 2018-12-28 | 2020-07-07 | 第四范式(北京)技术有限公司 | Method and system for recommending content |
CN111563158A (en) * | 2020-04-26 | 2020-08-21 | 腾讯科技(深圳)有限公司 | Text sorting method, sorting device, server and computer-readable storage medium |
CN111753184A (en) * | 2019-03-29 | 2020-10-09 | 北京达佳互联信息技术有限公司 | Information recommendation method and device |
CN111831922A (en) * | 2020-07-14 | 2020-10-27 | 深圳市众创达企业咨询策划有限公司 | Recommendation system and method based on internet information |
CN112667895A (en) * | 2020-12-28 | 2021-04-16 | 百果园技术(新加坡)有限公司 | Method, device and equipment for determining recommended item queue and storage medium |
CN112699157A (en) * | 2021-01-12 | 2021-04-23 | 腾讯科技(深圳)有限公司 | Information recommendation method, device, equipment and medium |
CN112800234A (en) * | 2021-04-15 | 2021-05-14 | 腾讯科技(深圳)有限公司 | Information processing method, device, electronic equipment and storage medium |
CN112995248A (en) * | 2019-12-12 | 2021-06-18 | 阿里巴巴集团控股有限公司 | Information pushing 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 |
CN113781146A (en) * | 2020-11-17 | 2021-12-10 | 北京沃东天骏信息技术有限公司 | Recommendation method, device, equipment and storage medium of product information |
CN115587261A (en) * | 2022-12-09 | 2023-01-10 | 思创数码科技股份有限公司 | Government affair resource catalog recommendation method and system |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100153370A1 (en) * | 2008-12-15 | 2010-06-17 | Microsoft Corporation | System of ranking search results based on query specific position bias |
US7921109B2 (en) * | 2005-10-05 | 2011-04-05 | Yahoo! Inc. | Customizable ordering of search results and predictive query generation |
CN102685565A (en) * | 2012-05-18 | 2012-09-19 | 合一网络技术(北京)有限公司 | Click feedback type individual recommendation system |
CN104090919A (en) * | 2014-06-16 | 2014-10-08 | 华为技术有限公司 | Advertisement recommending method and advertisement recommending server |
CN106504011A (en) * | 2015-09-06 | 2017-03-15 | 阿里巴巴集团控股有限公司 | A kind of methods of exhibiting of business object and device |
-
2018
- 2018-08-15 CN CN201810929455.XA patent/CN109086439B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7921109B2 (en) * | 2005-10-05 | 2011-04-05 | Yahoo! Inc. | Customizable ordering of search results and predictive query generation |
US20100153370A1 (en) * | 2008-12-15 | 2010-06-17 | Microsoft Corporation | System of ranking search results based on query specific position bias |
CN102685565A (en) * | 2012-05-18 | 2012-09-19 | 合一网络技术(北京)有限公司 | Click feedback type individual recommendation system |
CN104090919A (en) * | 2014-06-16 | 2014-10-08 | 华为技术有限公司 | Advertisement recommending method and advertisement recommending server |
CN106504011A (en) * | 2015-09-06 | 2017-03-15 | 阿里巴巴集团控股有限公司 | A kind of methods of exhibiting of business object and device |
Cited By (27)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111381909A (en) * | 2018-12-27 | 2020-07-07 | 北京字节跳动网络技术有限公司 | Page display method and device, terminal equipment and storage medium |
CN111381909B (en) * | 2018-12-27 | 2022-03-04 | 北京字节跳动网络技术有限公司 | Page display method and device, terminal equipment and storage medium |
CN111382346A (en) * | 2018-12-28 | 2020-07-07 | 第四范式(北京)技术有限公司 | Method and system for recommending content |
CN111382346B (en) * | 2018-12-28 | 2023-09-01 | 第四范式(北京)技术有限公司 | Method and system for recommending content |
CN111753184B (en) * | 2019-03-29 | 2024-01-02 | 北京达佳互联信息技术有限公司 | Information recommendation method and device |
CN111753184A (en) * | 2019-03-29 | 2020-10-09 | 北京达佳互联信息技术有限公司 | Information recommendation method and device |
CN109981785A (en) * | 2019-04-02 | 2019-07-05 | 北京字节跳动网络技术有限公司 | Method and apparatus for pushed information |
CN109981785B (en) * | 2019-04-02 | 2020-08-14 | 北京字节跳动网络技术有限公司 | Method and device for pushing information |
CN110110233A (en) * | 2019-05-09 | 2019-08-09 | 网易传媒科技(北京)有限公司 | Information processing method, device, medium and calculating equipment |
CN110659388B (en) * | 2019-10-10 | 2022-07-22 | 北京奇艺世纪科技有限公司 | To-be-recommended information screening method and device, electronic equipment and storage medium |
CN110659388A (en) * | 2019-10-10 | 2020-01-07 | 北京奇艺世纪科技有限公司 | To-be-recommended information screening method and device, electronic equipment and storage medium |
CN111008278B (en) * | 2019-11-22 | 2022-06-21 | 厦门美柚股份有限公司 | Content recommendation method and device |
CN110825974A (en) * | 2019-11-22 | 2020-02-21 | 厦门美柚股份有限公司 | Recommendation system content ordering method and device |
CN110825974B (en) * | 2019-11-22 | 2022-06-21 | 厦门美柚股份有限公司 | Recommendation system content ordering method and device |
CN111008278A (en) * | 2019-11-22 | 2020-04-14 | 厦门美柚股份有限公司 | Content recommendation method and device |
CN112995248A (en) * | 2019-12-12 | 2021-06-18 | 阿里巴巴集团控股有限公司 | Information pushing 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 |
CN111563158A (en) * | 2020-04-26 | 2020-08-21 | 腾讯科技(深圳)有限公司 | Text sorting method, sorting device, server and computer-readable storage medium |
CN111563158B (en) * | 2020-04-26 | 2023-08-29 | 腾讯科技(深圳)有限公司 | Text ranking method, ranking apparatus, server and computer-readable storage medium |
CN111831922A (en) * | 2020-07-14 | 2020-10-27 | 深圳市众创达企业咨询策划有限公司 | Recommendation system and method based on internet information |
CN113781146A (en) * | 2020-11-17 | 2021-12-10 | 北京沃东天骏信息技术有限公司 | Recommendation method, device, equipment and storage medium of product information |
CN112667895A (en) * | 2020-12-28 | 2021-04-16 | 百果园技术(新加坡)有限公司 | Method, device and equipment for determining recommended item queue and storage medium |
CN112699157A (en) * | 2021-01-12 | 2021-04-23 | 腾讯科技(深圳)有限公司 | Information recommendation method, device, equipment and medium |
CN112699157B (en) * | 2021-01-12 | 2023-10-20 | 腾讯科技(深圳)有限公司 | Information recommendation method, device, equipment and medium |
CN112800234B (en) * | 2021-04-15 | 2021-06-22 | 腾讯科技(深圳)有限公司 | Information processing method, device, electronic equipment and storage medium |
CN112800234A (en) * | 2021-04-15 | 2021-05-14 | 腾讯科技(深圳)有限公司 | Information processing method, device, electronic equipment and storage medium |
CN115587261A (en) * | 2022-12-09 | 2023-01-10 | 思创数码科技股份有限公司 | Government affair resource catalog recommendation method and system |
Also Published As
Publication number | Publication date |
---|---|
CN109086439B (en) | 2022-02-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109086439A (en) | Information recommendation method and device | |
CN110297848B (en) | Recommendation model training method, terminal and storage medium based on federal learning | |
WO2017181612A1 (en) | Personalized video recommendation method and device | |
CN108776676B (en) | Information recommendation method and device, computer readable medium and electronic device | |
US20180101540A1 (en) | Diversifying Media Search Results on Online Social Networks | |
CN110457581A (en) | A kind of information recommended method, device, electronic equipment and storage medium | |
US10949000B2 (en) | Sticker recommendation method and apparatus | |
CN109155136A (en) | The computerized system and method for highlight are detected and rendered automatically from video | |
CN109471978B (en) | Electronic resource recommendation method and device | |
CN111708876B (en) | Method and device for generating information | |
CN104782138A (en) | Identifying a thumbnail image to represent a video | |
US20210382609A1 (en) | Method and device for displaying multimedia resource | |
CN110413867B (en) | Method and system for content recommendation | |
US20170351769A1 (en) | System and Method for a Platform to Identify and Connect Like-Minded Individuals Based on Interaction | |
CN109168047A (en) | Video recommendation method, device, server and storage medium | |
CN112000871A (en) | Method, device and equipment for determining search result list and storage medium | |
CN112818224A (en) | Information recommendation method and device, electronic equipment and readable storage medium | |
CN111552835B (en) | File recommendation method, device and server | |
CN107968743A (en) | The method and apparatus of pushed information | |
CN112559777A (en) | Content item delivery method and device, computer equipment and storage medium | |
CN115935049A (en) | Recommendation processing method and device based on artificial intelligence and electronic equipment | |
CN112287799A (en) | Video processing method and device based on artificial intelligence and electronic equipment | |
CN111475664A (en) | Object display method and device and electronic equipment | |
CN111414966B (en) | Classification method, classification device, electronic equipment and computer storage medium | |
CN112533050B (en) | Video processing method, device, equipment and medium |
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 |