CN108334575A - A kind of recommendation results sequence modification method and device, electronic equipment - Google Patents
A kind of recommendation results sequence modification method and device, electronic equipment Download PDFInfo
- Publication number
- CN108334575A CN108334575A CN201810064755.6A CN201810064755A CN108334575A CN 108334575 A CN108334575 A CN 108334575A CN 201810064755 A CN201810064755 A CN 201810064755A CN 108334575 A CN108334575 A CN 108334575A
- Authority
- CN
- China
- Prior art keywords
- recommendation results
- probability distribution
- real
- recommendation
- distribution model
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Abstract
This application involves a kind of recommendation results sequence modification methods, belong to field of computer technology, solve the problems, such as to recommend recommendation results accuracy rate present in sort method low in the prior art.Disclosed in the embodiment of the present application recommendation results sequence modification method include:Obtain real-time behavioral data of the active user to history recommendation results, according to the real-time behavioral data, construct the probability Distribution Model that the active user executes the behavior of presetting for current recommendation results, and further the current recommendation results are modified according to the sampled value of the probability Distribution Model.The sequence modification method of the application is modified the sequence of recommendation results list by the real-time behavioral statistics amount based on user, and behavioural habits of the recommendation results based on user obtain, and effectively improve the accuracy rate of recommendation results, meanwhile, the user experience is improved.
Description
Technical field
This application involves field of computer technology, more particularly to a kind of recommendation results sequence modification method and device, electricity
Sub- equipment.
Background technology
Personalized ordering for user is a critically important research topic of commending system the inside.With network platform industry
The fast development of business, each user have businessmans up to a hundred and product to be called back, enter at the exhibition position on the access platform page
Candidate Set shows user.In the prior art, recommendation results list is mainly by recommended models, such as Learn to rank (sequences
Study) model acquisition.Machine learning techniques are applied in sequence by Learn to rank models, by being pushed away to each user
It recommends sort result problem and is abstracted as optimization problem;Data source is the User action log in the past period, passes through feature
Engineering construction sequencing feature solves optimization problem.This sort method is limited to the User action log of system acquisition, because
On the overdue or line of user the problem of log recording, many noises are had in the daily record that can cause;Behavior-based control daily record
Training machine learning model, it will cause the model learnt inaccurate, in order to promote recommendation effect, it is common practice that recommending
The sequence of the results list is modified.
Modification method commonly used in the prior art includes:Human intervention is carried out for recommendation results list, it is multiple to exposing,
But the recommendation results that do not click always do the processing of drop power;Alternatively, increasing the friendship of user-recommendation results dimension in training data
Pitch feature.However, temporary based on clicking rate tune, in the case where recommendation results exposure frequency is enough, clicking rate is one normal
Amount, the i.e. each tune weights of recommendation results are fixed, and do not have the effect of optimization recommendation results.And user-recommendation results
The more difficult measurement of cross feature of dimension, equally can not effectively correct the sequence of recommendation results.
The modification method as it can be seen that recommendation results in the prior art sort, it is low can not still to solve recommendation results accuracy rate
Problem.
Invention content
The application provides a kind of recommendation results sequence modification method, at least solves recommendation sort method in the prior art and deposits
The low problem of recommendation results accuracy rate.
To solve the above-mentioned problems, in a first aspect, the embodiment of the present application provides a kind of recommendation results sequence modification method
Including:
Obtain real-time behavioral data of the active user to history recommendation results;
According to the real-time behavioral data, constructs the active user and execute the general of the behavior of presetting for current recommendation results
Rate distributed model;
According to the sampled value of the probability Distribution Model, the current recommendation results are modified.
Second aspect, the embodiment of the present application provide a kind of recommendation results sequence correcting device, including:
Real-time behavioral data acquisition module, for obtaining real-time behavioral data of the active user to history recommendation results;
User behavior probability Distribution Model constructing module, for according to the real-time behavioral data, constructing the current use
Family executes the probability Distribution Model for the behavior of presetting for current recommendation results;
Sort correcting module, for the sampled value according to the probability Distribution Model, is carried out to the current recommendation results
It corrects.
The third aspect, the embodiment of the present application also disclose a kind of electronic equipment, including memory, processor and are stored in institute
The computer program that can be run on memory and on a processor is stated, the processor realizes this when executing the computer program
Apply for the recommendation results sequence modification method described in embodiment.
Fourth aspect, the embodiment of the present application provide a kind of computer readable storage medium, are stored thereon with computer journey
Sequence, when which is executed by processor disclosed in the embodiment of the present application the step of recommendation results sequence modification method.
Recommendation results sequence modification method disclosed in the embodiment of the present application, by obtaining active user to history recommendation results
Real-time behavioral data then according to the real-time behavioral data, construct the active user and executed for current recommendation results
The probability Distribution Model of default behavior, and further the current recommendation is tied according to the sampled value of the probability Distribution Model
Fruit is modified, and solves the problems, such as to recommend recommendation results accuracy rate existing for sort method low in the prior art.The application base
The sequence of recommendation results list is modified in the real-time behavioral statistics amount of user, behavioural habits of the recommendation results based on user
It obtains, effectively improves the accuracy rate of recommendation results, meanwhile, the user experience is improved.
Description of the drawings
It, below will be in embodiment or description of the prior art in order to illustrate more clearly of the technical solution of the embodiment of the present application
Required attached drawing is briefly described, it should be apparent that, the accompanying drawings in the following description is only some realities of the application
Example is applied, it for those of ordinary skill in the art, without having to pay creative labor, can also be attached according to these
Figure obtains other attached drawings.
Fig. 1 is the recommendation results sequence modification method flow chart of the embodiment of the present application one;
Fig. 2 is a Beta distribution schematic diagram of the recommendation results sequence modification method construction of the embodiment of the present application two;
Fig. 3 is the structural schematic diagram of the recommendation results sequence correcting device of the embodiment of the present application three.
Specific implementation mode
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Site preparation describes, it is clear that described embodiment is some embodiments of the present application, instead of all the embodiments.Based on this Shen
Please in embodiment, the every other implementation that those of ordinary skill in the art are obtained without creative efforts
Example, shall fall in the protection scope of this application.
Embodiment one
A kind of recommendation results sequence modification method disclosed in the present embodiment, as shown in Figure 1, this method includes:Step 110 to
Step 130.
Step 110, real-time behavioral data of the active user to history recommendation results is obtained.
When it is implemented, parsing active user to shown recommendation results from the real-time behavioral data stream of the whole network user
Default behavior list, it is for statistical analysis to the behavior list of the active user then from the category dimension of recommendation results,
User is determined in past designated time period (such as two hours), to the recommendation results of the different categories of exposed (showing)
Number of clicks and non-number of clicks.It is determined according to the optimization aim of recommendation results when it is implemented, presetting behavior herein.Example
Such as, for improve commending system of the clicking rate as optimization aim, the default behavior includes " click ";For being visited with improving
Purchase rate is the commending system of optimization aim, and the default behavior includes " purchase ".
When it is implemented, can also be for statistical analysis to the behavior list of the active user from recommendation results dimension,
User is determined in past designated time period (such as two hours), the click to the different recommendation results of exposed (showing)
Number and non-number of clicks.
Step 120, according to the real-time behavioral data, it is default for the execution of current recommendation results to construct the active user
The probability Distribution Model of behavior.
When it is implemented, the real-time behavioral data includes:User is to having exposed the default behavior numbers of recommendation results.With
Default behavior is for " click ", the real-time behavioral data includes:User is in past preset time (such as in two hours)
To the number of clicks of the recommendation results of exposure, and number is not clicked on.Then, using the number of clicks and do not click on number as join
Number, structuring user's execute current recommendation results the probability Distribution Model of click behavior.When it is implemented, the probability distribution mould
Type can be Beta distributed models, or Gaussian distribution model.
Step 130, according to the sampled value of the probability Distribution Model, the current recommendation results are modified.
After constructing probability Distribution Model, by being sampled to probability distribution curve, using obtained sampled value as
Weight, corrects the recommendation scores of recommendation results in recommendation results list, and further resequences to recommendation results list,
To play the role of correcting recommendation results.
Recommendation results sequence modification method disclosed in the embodiment of the present application, by obtaining active user to history recommendation results
Real-time behavioral data then according to the real-time behavioral data, construct the active user and executed for current recommendation results
The probability Distribution Model of default behavior, and further the current recommendation is tied according to the sampled value of the probability Distribution Model
Fruit is modified, and solves the problems, such as to recommend recommendation results accuracy rate existing for sort method low in the prior art.The application base
The sequence of recommendation results list is modified in the real-time behavioral statistics amount of user, behavioural habits of the recommendation results based on user
It obtains, effectively improves the accuracy rate of recommendation results, meanwhile, the user experience is improved.
Embodiment two
The present embodiment is a specific embodiment of recommendation results sequence modification method disclosed in the present application,.
Based on embodiment one, when it is implemented, described obtain real-time behavioral data of the active user to history recommendation results,
Including:Obtain the real-time behavioral data performed by the three-level category recommendation results that active user exposes history.It is current obtaining
When user is to the real-time behavioral datas of history recommendation results, real time data can be obtained from the dimension of category, it can also be from product
Dimension obtains real time data.Since commending system product is numerous, real time data is obtained from product dimension, it is dilute there may be data
Thin problem, i.e., within a preset period of time, certain products may be never exposed, and system will be unable to obtain user to the product
Real-time stream.It is preferred, therefore, that being based on category dimension, obtain performed by the recommendation results that active user exposes history
Real-time behavioral data.When it is implemented, the grade label of different stage can be arranged in the product of commending system.For example, recommendation results
The primes class label of " ports the Jin Quan worlds I MAX film city " is " amusement and recreation ", and seconds class label is " leisure ", three-level category
Label is " cinema ", and level Four category label is " Chaoyang District ".As it can be seen that I and II category is too wide in range, not careful enough, user couple
Real-time behavioral data performed by the I and II category recommendation results of history exposure can not embody user to different recommendation results
Discrimination.And level Four and following rank category are excessively careful, and real time data is obtained from level Four and following rank category dimension, it may
Can there are problems that Sparse.Preferably, the application obtains the three-level category recommendation results institute that active user exposes history
The real-time behavioral data executed.
When it is implemented, the real-time behavioral data includes:In designated time period, to each three-level category of history exposure
Under recommendation results execute the number for the behavior of presetting and do not execute the number of default behavior.
The application when it is implemented, the default behavior according to the optimization aim of recommendation results determine, the default behavior
Including but not limited to:It clicks or buys.If commending system presets row using the clicking rate for promoting recommendation results as optimization aim
It is to include:Click recommendation results;If commending system presets behavior packet using the visit purchase rate for promoting recommendation results as optimization aim
It includes:Buy recommendation results.
In the following, being to click to illustrate with default behavior, detailed description obtains active user and pushed away to the three-level category that history exposes
Recommend the specific technical solution of the real-time behavioral data performed by result.
First, from all recommendation results exposed to active user in acquisition designated time period in the whole network real time data.
The designated time period determines according to specific requirements, such as can be nearest two hours.Then, according to the three-level of each recommendation results
Category label classifies to exposed recommendation results, and the corresponding recommendation results of each three-level category include multiple.Specifically
When implementation, the total exposure number of all recommendation results under the corresponding classification of each three-level category is recommended to tie as the three-level category
Total exposure number of the fruit to active user;Total point of all recommendation results under some corresponding classification of three-level category of active user
Hit total number of clicks of the number as active user to the three-level category recommendation results;Each three-level category is to the total of active user
Exposure frequency subtracts total number of clicks of the active user to the three-level category recommendation results, obtains active user to the three-level category
The exposure of recommendation results does not click on number.Assuming that recommendation results exposed in nearest two hours include:C1,C2,C3Three three
Grade category, wherein category C1Include recommendation results p1And p2If recommendation results p1It has exposed 50 times, active user clicks
10 times, recommendation results p250 times are exposed, then three-level category C1Active user is exposed in total 100 times, active user clicks 10
It is secondary.According to the method, can obtain in designated time period, each three-level category recommendation results exposed to active user
Exposure does not click on number and exposure number of clicks.
When it is implemented, can also be modeled by the method (such as maximum likelihood) of machine learning, in conjunction with the reality of active user
When behavioral data, as training data carry out parameter learning, obtain the parameter of probability Distribution Model.
Then, it according to the real-time behavioral data of the active user of acquisition, constructs the active user and is directed to and ought be pushed forward
Recommend the probability Distribution Model that result executes click behavior.When it is implemented, according to the real-time behavioral data, construct described current
User for current recommendation results execute preset behavior probability Distribution Model include:History is exposed according to active user every
The number of the default behavior of recommendation results execution under a three-level category and the number for not executing default behavior determine model parameter, structure
Make the probability Distribution Model that the active user executes the behavior of presetting for corresponding three-level category recommendation results.
When it is implemented, not clicked on according to the exposure of each three-level category recommendation results exposed to active user secondary
Number and exposure number of clicks construct the probability Distribution Model that the active user executes click behavior for current recommendation results.Its
In, probability Distribution Model can be that beta is distributed (Beta Distribution) distributed model, or Gaussian Profile mould
Type.In the present embodiment, for building Beta distributed models, it is described in detail according to each three-level exposed to active user
The exposure of category recommendation results does not click on number and exposure number of clicks constructs the active user and held for current recommendation results
The probability Distribution Model of row click behavior.
Beta distribution (Beta distributions) model is a conjugate gradient descent method as Bernoulli Jacob's distribution and binomial distribution
Density function, have important application in machine learning and mathematical statistics.In probability theory, Beta distributions refer to one group of definition
Continuous probability distribution in (0,1) section.Beta distributed models are there are two parameter, respectively α and β, in the present embodiment, another α=
γ+exposure number of clicks, β=γ+exposure do not click on number, wherein γ is smaller constant, such as value is 1.
With active user to exposed three-level category C1Recommendation results exposure do not click on number and exposure click
Number constructs the active user for belonging to three-level category C in current recommendation results1Recommendation results execute click behavior
Beta distributed models are illustrated, and α=1+10, β=1+90 can then obtain Beta distribution curves as shown in Figure 2.
Beta is distributed, mean value E=α/(alpha+beta) of α and β, in the present embodiment, the mean value E=11/102 of Beta distributions, and
Be worth it is smaller, i.e., to Beta profile samples shown in Fig. 2 when, have the sampled value that very maximum probability obtains can be smaller, therefore, with
The sampled value of Beta distributions is adjusted recommendation scores as weight, will play and suppresses the corresponding three superfine product class of Beta distributions
The effect of recommendation results sequence.As it can be seen that the size of the sampled value of Beta distributions recently recommends the three-level category with active user
As a result exposure clicking rate positive correlation, has fully demonstrated the search need of user.
According to the method described above, click behavior probability distributed mode of the user to each three-level category recommendation results can be obtained
Type.If in the real-time behavioral data performed by some three-level category recommendation results that the active user obtained exposes history,
Exposure number of clicks is relatively more, and the mean value of Beta distributions can be bigger, and the sampled value meeting maximum probability of Beta distributions is fallen in mean value E
Near, therefore the sampled value obtained this when can be larger, using the sampled value that Beta is distributed as weight to recommendation scores into
Row adjustment, will play the role of promoting the Beta and is distributed corresponding three superfine product class and recommends sort result.
When it is implemented, if commending system visits purchase rate as optimization aim, according to the active user couple of acquisition to be promoted
Real-time buying behavior data performed by the three-level category recommendation results of history exposure, structuring user's recommend each three-level category
As a result default behavior probability distributed model.
When it is implemented, Gaussian distribution model, the present embodiment can also be constructed according to user's real-time behavioral data of acquisition
It repeats no more.
After obtaining user to the default behavior probability distributed model of each three-level category recommendation results, further basis
The sampled value of the probability Distribution Model is modified corresponding recommendation results in the current recommendation results.
When it is implemented, the sampled value according to the probability Distribution Model, to the current recommendation results amendment, packet
It includes:By carrying out stochastical sampling to the corresponding probability distribution curve of the probability Distribution Model, the probability Distribution Model is determined
Sampled value;By the product of the sampled value and the recommendation scores of the current recommendation results, as the current recommendation results
Revised recommendation scores;According to the revised recommendation scores, the sequence of the current recommendation results is modified.
After the default behavior probability distributed model for constructing each three-level category recommendation results, according to the sampled value of each probability distribution,
The recommendation scores of the recommendation results to belonging to corresponding three-level category are modified respectively.
Include recommendation results p with the recommendation results list that commending system returns1For, because of recommendation results p1Three-level product
Class label is C1, therefore, by three-level category recommendation results C1The corresponding probability distribution curve of probability Distribution Model adopted
Sample obtains sampled value sample (Beta (C1)), then, by the sampled value and commending system to recommendation results p1That estimates pushes away
Recommend score (p1) product, as current recommendation results p1Consequently recommended score.That is, passing through formula ctr_new (p1)=ctr
(p1)*sample(Beta(C1)), calculate recommendation results p1Consequently recommended score, to commending system calculate sequence score into
Row is corrected.If the recommendation results list that commending system returns includes recommendation results p3, and recommendation results p3Three-level category label
For C2, by three-level category recommendation results C2The corresponding probability distribution curve of probability Distribution Model sampled, sampled
Value sample (Beta (C2)), and pass through formula ctr_new (p3)=ctr (p3)*sample(Beta(C2)), it calculates and recommends knot
Fruit p3Consequently recommended score, wherein ctr (p3) be commending system to recommendation results p3The recommendation scores estimated.
According to the method described above, it can obtain in the recommendation results list of commending system return after the amendment of each recommendation results
Recommendation scores.Finally, it resequences to recommendation results according to revised recommendation scores, and exports to client and carry out
Displaying.
This programme is based on user's three-level category behavioral statistics amount in user's real-time stream, in conjunction with Beta profile samples,
The recommendation scores estimated to recommended models in commending system intervene in real time and are corrected, and it is big significantly to suppress light exposure, point
The displaying of the few recommendation results of number is hit, promotes user experience, while promoting day any active ues quantity.It is distributed to obtain using Beta
Random value is modified recommendation scores, can promote the diversity of the recommendation results of commending system return.
Recommendation results sequence modification method disclosed in the embodiment of the present application, by obtaining active user to history recommendation results
Real-time behavioral data then according to the real-time behavioral data, construct the active user and executed for current recommendation results
The probability Distribution Model of default behavior, and further the current recommendation is tied according to the sampled value of the probability Distribution Model
Fruit is modified, and solves the problems, such as to recommend recommendation results accuracy rate existing for sort method low in the prior art.The application base
The sequence of recommendation results list is modified in the real-time behavioral statistics amount of user, behavioural habits of the recommendation results based on user
It obtains, effectively improves the accuracy rate of recommendation results, meanwhile, the user experience is improved.
By in three-level category using Beta distribution recommendation scores are intervened, for user click frequency it is high three
Grade category, can come before recommendation results list after intervention, increase chance for exposure;For user click frequency it is few three
Grade category, can come behind recommendation results list after intervention, reduce chance for exposure, and then improve the standard of recommendation results
Exactness.Meanwhile if directly obtained in the real-time behavioral data of recommendation results dimension, it is secondary so that the past period user clicks
The more recommendation results of number expose always, and user experience is bad, using three-level category, can increase with user click frequency more than recommendation
As a result the exposure frequency for belonging to the recommendation results of a three-level category recommends diversity helpful to being promoted.
Embodiment three
A kind of recommendation results sequence correcting device disclosed in the present embodiment, as shown in figure 3, described device includes:
Real-time behavioral data acquisition module 310, for obtaining real-time behavioral data of the active user to history recommendation results;
User behavior probability Distribution Model constructing module 320, for according to the real-time behavioral data, constructing described current
User executes the probability Distribution Model for the behavior of presetting for current recommendation results;
Sort correcting module 330, for according to the sampled value of the probability Distribution Model, to the current recommendation results into
Row is corrected.
Optionally, the real-time behavioral data acquisition module 310, is further used for:
Obtain the real-time behavioral data performed by the three-level category recommendation results that active user exposes history.
Optionally, the real-time behavioral data includes:In designated time period, under each three-level category of history exposure
Recommendation results execute the number for the behavior of presetting and do not execute the number of default behavior.
When it is implemented, probability Distribution Model can be obtained by the method for statistics according to user's real-time behavioral data
Parameter;It can also be modeled by the method (such as maximum likelihood) of machine learning, in conjunction with the real-time behavioral data of active user, as
Training data carries out parameter learning, obtains the parameter of probability Distribution Model, the application does not limit this.
Optionally, the user behavior probability Distribution Model constructing module 320, is further used for:
Recommendation results under each three-level category exposed to history according to active user execute the behavior of presetting number and
The number for not executing default behavior determines model parameter, constructs the active user and is executed for corresponding three-level category recommendation results
The probability Distribution Model of default behavior.
When it is implemented, the probability Distribution Model can be Beta distributed models, or Gaussian distribution model.
Optionally, the sequence correcting module 330 is further used for:
By carrying out stochastical sampling to the corresponding probability distribution curve of the probability Distribution Model, the probability distribution is determined
The sampled value of model;
By the product of the sampled value and the recommendation scores of the current recommendation results, repaiied as the current recommendation results
Recommendation scores after just;
According to the revised recommendation scores, the sequence of the current recommendation results is modified.
This programme builds probability Distribution Model, knot based on user's three-level category behavioral statistics amount in user's real-time stream
The sampled value for closing probability distribution to the recommendation scores that recommended models in commending system are estimated in real time intervene and be corrected, Ke Yi great
Amplitude suppresses that light exposure is big, the displaying of the few recommendation results of number of clicks, promotes user experience, while promoting a day active users
Amount.Recommendation scores are modified using probability distribution stochastical sampling value, the recommendation results of commending system return can be promoted
Diversity.
Optionally, the default behavior is determined according to the optimization aim of recommendation results, and the default behavior includes:Click or
Purchase.
Recommendation results sequence correcting device disclosed in the embodiment of the present application, by obtaining active user to history recommendation results
Real-time behavioral data then according to the real-time behavioral data, construct the active user and executed for current recommendation results
The probability Distribution Model of default behavior, and further the current recommendation is tied according to the sampled value of the probability Distribution Model
Fruit is modified, and solves the problems, such as to recommend recommendation results accuracy rate existing for sort method low in the prior art.The application base
The sequence of recommendation results list is modified in the real-time behavioral statistics amount of user, behavioural habits of the recommendation results based on user
It obtains, effectively improves the accuracy rate of recommendation results, meanwhile, the user experience is improved.
By in three-level category using Beta distribution recommendation scores are intervened, for user click frequency it is high three
Grade category, can come before recommendation results list after intervention, increase chance for exposure;For user click frequency it is few three
Grade category, can come behind recommendation results list after intervention, reduce chance for exposure, and then improve the standard of recommendation results
Exactness.Meanwhile if directly obtained in the real-time behavioral data of recommendation results dimension, it is secondary so that the past period user clicks
The more recommendation results of number expose always, and user experience is bad, using three-level category, can increase with user click frequency more than recommendation
As a result the exposure frequency for belonging to the recommendation results of a three-level category recommends diversity helpful to being promoted.
Correspondingly, disclosed herein as well is a kind of electronic equipment, including memory, processor and it is stored in the memory
Computer program that is upper and can running on a processor, the processor are realized when executing the computer program as the application is real
Apply the recommendation results sequence modification method described in example one and embodiment two.The electronic equipment can be PC machine, mobile terminal, a
Personal digital assistant, tablet computer etc..
Disclosed herein as well is a kind of computer readable storage mediums, are stored thereon with computer program, which is located
Manage the step of recommendation results sequence modification method as described in the embodiment of the present application one and embodiment two is realized when device executes.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with
The difference of other embodiment, the same or similar parts between the embodiments can be referred to each other.For device embodiment
For, since it is basically similar to the method embodiment, so description is fairly simple, referring to the portion of embodiment of the method in place of correlation
It defends oneself bright.
A kind of recommendation results sequence modification method provided by the present application and device are described in detail above, herein
Applying specific case, the principle and implementation of this application are described, and the explanation of above example is only intended to help
Understand the present processes and its core concept;Meanwhile for those of ordinary skill in the art, according to the thought of the application,
There will be changes in the specific implementation manner and application range, in conclusion the content of the present specification should not be construed as to this
The limitation of application.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It is realized by the mode of software plus required general hardware platform, naturally it is also possible to pass through hardware realization.Based on such reason
Solution, substantially the part that contributes to existing technology can embody above-mentioned technical proposal in the form of software products in other words
Come, which can store in a computer-readable storage medium, such as ROM/RAM, magnetic disc, CD, including
Some instructions are used so that a computer equipment (can be personal computer, server or the network equipment etc.) executes respectively
Method described in certain parts of a embodiment or embodiment.
Claims (10)
- The modification method 1. a kind of recommendation results sort, which is characterized in that including:Obtain real-time behavioral data of the active user to history recommendation results;According to the real-time behavioral data, the probability point that the active user executes the behavior of presetting for current recommendation results is constructed Cloth model;According to the sampled value of the probability Distribution Model, the current recommendation results are modified.
- 2. according to the method described in claim 1, it is characterized in that, the acquisition active user is to the real-time of history recommendation results The step of behavioral data, including:Obtain the real-time behavioral data performed by the three-level category recommendation results that active user exposes history.
- 3. according to the method described in claim 2, it is characterized in that, the real-time behavioral data includes:It is right in designated time period Recommendation results under each three-level category of history exposure execute the number for the behavior of presetting and do not execute the number of default behavior.
- 4. according to the method described in claim 3, it is characterized in that, described according to the real-time behavioral data, work as described in construction Preceding user executes the step of probability Distribution Model for the behavior of presetting for current recommendation results, including:Recommendation results under each three-level category exposed to history according to active user execute the number for the behavior of presetting and do not hold The number of the default behavior of row determines model parameter, and it is default for the execution of corresponding three-level category recommendation results to construct the active user The probability Distribution Model of behavior.
- 5. method according to any one of claims 1 to 4, which is characterized in that described according to the probability Distribution Model Sampled value, the step of being modified to the current recommendation results, including:By carrying out stochastical sampling to the corresponding probability distribution curve of the probability Distribution Model, the probability Distribution Model is determined Sampled value;By the product of the sampled value and the recommendation scores of the current recommendation results, after being corrected as the current recommendation results Recommendation scores;According to the revised recommendation scores, the sequence of the current recommendation results is modified.
- 6. method according to any one of claims 1 to 4, which is characterized in that the default behavior is according to recommendation results Optimization aim determines that the default behavior includes:It clicks or buys.
- The correcting device 7. a kind of recommendation results sort, which is characterized in that including:Real-time behavioral data acquisition module, for obtaining real-time behavioral data of the active user to history recommendation results;User behavior probability Distribution Model constructing module, for according to the real-time behavioral data, constructing active user's needle Current recommendation results are executed with the probability Distribution Model for the behavior of presetting;The correcting module that sorts is modified the current recommendation results for the sampled value according to the probability Distribution Model.
- 8. device according to claim 7, which is characterized in that the sequence correcting module is further used for:By carrying out stochastical sampling to the corresponding probability distribution curve of the probability Distribution Model, the probability Distribution Model is determined Sampled value;By the product of the sampled value and the recommendation scores of the current recommendation results, after being corrected as the current recommendation results Recommendation scores;According to the revised recommendation scores, the sequence of the current recommendation results is modified.
- 9. a kind of electronic equipment, including memory, processor and it is stored on the memory and can runs on a processor Computer program, which is characterized in that the processor realizes claim 1 to 6 any one when executing the computer program The recommendation results sequence modification method.
- 10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor The step of recommendation results sequence modification method described in claim 1 to 6 any one is realized when execution.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810064755.6A CN108334575B (en) | 2018-01-23 | 2018-01-23 | Recommendation result sorting correction method and device and electronic equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810064755.6A CN108334575B (en) | 2018-01-23 | 2018-01-23 | Recommendation result sorting correction method and device and electronic equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108334575A true CN108334575A (en) | 2018-07-27 |
CN108334575B CN108334575B (en) | 2022-04-26 |
Family
ID=62926203
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810064755.6A Active CN108334575B (en) | 2018-01-23 | 2018-01-23 | Recommendation result sorting correction method and device and electronic equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108334575B (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109257648A (en) * | 2018-09-30 | 2019-01-22 | 武汉斗鱼网络科技有限公司 | A kind of direct broadcasting room similarity modification method, device, terminal and storage medium |
CN109558544A (en) * | 2018-12-12 | 2019-04-02 | 拉扎斯网络科技(上海)有限公司 | Sort method and device, server and storage medium |
CN109597941A (en) * | 2018-12-12 | 2019-04-09 | 拉扎斯网络科技(上海)有限公司 | Sort method and device, electronic equipment and storage medium |
CN110347781A (en) * | 2019-07-18 | 2019-10-18 | 腾讯科技(深圳)有限公司 | Article falls discharge method, article recommended method, device, equipment and storage medium |
CN111275493A (en) * | 2020-02-10 | 2020-06-12 | 拉扎斯网络科技(上海)有限公司 | List data processing method and device, server and nonvolatile storage medium |
CN111382349A (en) * | 2018-12-29 | 2020-07-07 | 广州市百果园网络科技有限公司 | Information recommendation method and device, computer equipment and 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 |
CN111612581A (en) * | 2020-05-18 | 2020-09-01 | 深圳市分期乐网络科技有限公司 | Method, device and equipment for recommending articles and storage medium |
CN111782927A (en) * | 2019-05-15 | 2020-10-16 | 北京京东尚科信息技术有限公司 | Article recommendation method and device, computer storage medium |
CN112231593A (en) * | 2020-12-15 | 2021-01-15 | 上海朝阳永续信息技术股份有限公司 | Financial information intelligent recommendation system |
CN113190758A (en) * | 2021-05-21 | 2021-07-30 | 聚好看科技股份有限公司 | Server and media asset recommendation method |
CN114037486A (en) * | 2022-01-07 | 2022-02-11 | 阿里健康科技(中国)有限公司 | Method for determining user appeal, application bearing method and device |
CN115083442A (en) * | 2022-04-29 | 2022-09-20 | 马上消费金融股份有限公司 | Data processing method, data processing device, electronic equipment and computer readable storage medium |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102004782A (en) * | 2010-11-25 | 2011-04-06 | 北京搜狗科技发展有限公司 | Search result sequencing method and search result sequencer |
US20110161330A1 (en) * | 2007-04-30 | 2011-06-30 | Microsoft Corporation | Calculating global importance of documents based on global hitting times |
US20110213786A1 (en) * | 2010-02-26 | 2011-09-01 | International Business Machines Corporation | Generating recommended items in unfamiliar domain |
CN102193999A (en) * | 2011-05-09 | 2011-09-21 | 北京百度网讯科技有限公司 | Method and device for sequencing search results |
CN104794135A (en) * | 2014-01-21 | 2015-07-22 | 阿里巴巴集团控股有限公司 | Method and device for carrying out sorting on search results |
CN104866474A (en) * | 2014-02-20 | 2015-08-26 | 阿里巴巴集团控股有限公司 | Personalized data searching method and device |
CN105740444A (en) * | 2016-02-02 | 2016-07-06 | 桂林电子科技大学 | User score-based project recommendation method |
EP3139284A1 (en) * | 2015-09-01 | 2017-03-08 | Dream It Get IT Limited | Media unit retrieval and related processes |
CN107368519A (en) * | 2017-06-05 | 2017-11-21 | 桂林电子科技大学 | A kind of cooperative processing method and system for agreeing with user interest change |
-
2018
- 2018-01-23 CN CN201810064755.6A patent/CN108334575B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110161330A1 (en) * | 2007-04-30 | 2011-06-30 | Microsoft Corporation | Calculating global importance of documents based on global hitting times |
US20110213786A1 (en) * | 2010-02-26 | 2011-09-01 | International Business Machines Corporation | Generating recommended items in unfamiliar domain |
CN102004782A (en) * | 2010-11-25 | 2011-04-06 | 北京搜狗科技发展有限公司 | Search result sequencing method and search result sequencer |
CN102193999A (en) * | 2011-05-09 | 2011-09-21 | 北京百度网讯科技有限公司 | Method and device for sequencing search results |
CN104794135A (en) * | 2014-01-21 | 2015-07-22 | 阿里巴巴集团控股有限公司 | Method and device for carrying out sorting on search results |
CN104866474A (en) * | 2014-02-20 | 2015-08-26 | 阿里巴巴集团控股有限公司 | Personalized data searching method and device |
EP3139284A1 (en) * | 2015-09-01 | 2017-03-08 | Dream It Get IT Limited | Media unit retrieval and related processes |
CN105740444A (en) * | 2016-02-02 | 2016-07-06 | 桂林电子科技大学 | User score-based project recommendation method |
CN107368519A (en) * | 2017-06-05 | 2017-11-21 | 桂林电子科技大学 | A kind of cooperative processing method and system for agreeing with user interest change |
Non-Patent Citations (2)
Title |
---|
ROHIT MALGAONKAR 等: ""Image re-ranking semantic search engine : Reinforcement learning methodology"", 《2016 INTERNATIONAL CONFERENCE ON ELECTRICAL, ELECTRONICS, AND OPTIMIZATION TECHNIQUES (ICEEOT)》 * |
洪宇 等: ""一种新型最优检索结果的发现与论证"", 《计算机学报》 * |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109257648A (en) * | 2018-09-30 | 2019-01-22 | 武汉斗鱼网络科技有限公司 | A kind of direct broadcasting room similarity modification method, device, terminal and storage medium |
CN109257648B (en) * | 2018-09-30 | 2021-03-16 | 武汉斗鱼网络科技有限公司 | Method, device, terminal and storage medium for correcting similarity between live broadcasts |
CN109558544A (en) * | 2018-12-12 | 2019-04-02 | 拉扎斯网络科技(上海)有限公司 | Sort method and device, server and storage medium |
CN109597941A (en) * | 2018-12-12 | 2019-04-09 | 拉扎斯网络科技(上海)有限公司 | Sort method and device, electronic equipment and storage medium |
CN109558544B (en) * | 2018-12-12 | 2021-04-27 | 拉扎斯网络科技(上海)有限公司 | Sorting method and device, server and storage medium |
CN111382349A (en) * | 2018-12-29 | 2020-07-07 | 广州市百果园网络科技有限公司 | Information recommendation method and device, computer equipment and storage medium |
CN111382349B (en) * | 2018-12-29 | 2023-08-15 | 广州市百果园网络科技有限公司 | Information recommendation method, device, computer equipment and storage medium |
CN111782927A (en) * | 2019-05-15 | 2020-10-16 | 北京京东尚科信息技术有限公司 | Article recommendation method and device, computer storage medium |
CN110347781A (en) * | 2019-07-18 | 2019-10-18 | 腾讯科技(深圳)有限公司 | Article falls discharge method, article recommended method, device, equipment and storage medium |
CN110347781B (en) * | 2019-07-18 | 2023-10-20 | 深圳市雅阅科技有限公司 | Article reverse arrangement method, article recommendation method, device, equipment and storage medium |
CN111275493A (en) * | 2020-02-10 | 2020-06-12 | 拉扎斯网络科技(上海)有限公司 | List data processing method and device, server and nonvolatile storage medium |
CN111275493B (en) * | 2020-02-10 | 2023-08-22 | 拉扎斯网络科技(上海)有限公司 | Processing method and device of list data, server and nonvolatile 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 |
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 |
CN111612581A (en) * | 2020-05-18 | 2020-09-01 | 深圳市分期乐网络科技有限公司 | Method, device and equipment for recommending articles and storage medium |
CN112231593A (en) * | 2020-12-15 | 2021-01-15 | 上海朝阳永续信息技术股份有限公司 | Financial information intelligent recommendation system |
CN113190758A (en) * | 2021-05-21 | 2021-07-30 | 聚好看科技股份有限公司 | Server and media asset recommendation method |
CN113190758B (en) * | 2021-05-21 | 2023-01-20 | 聚好看科技股份有限公司 | Server and media asset recommendation method |
CN114037486A (en) * | 2022-01-07 | 2022-02-11 | 阿里健康科技(中国)有限公司 | Method for determining user appeal, application bearing method and device |
CN115083442B (en) * | 2022-04-29 | 2023-08-08 | 马上消费金融股份有限公司 | Data processing method, device, electronic equipment and computer readable storage medium |
CN115083442A (en) * | 2022-04-29 | 2022-09-20 | 马上消费金融股份有限公司 | Data processing method, data processing device, electronic equipment and computer readable storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN108334575B (en) | 2022-04-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108334575A (en) | A kind of recommendation results sequence modification method and device, electronic equipment | |
US11659050B2 (en) | Discovering signature of electronic social networks | |
CN110033314B (en) | Advertisement data processing method and device | |
US8775332B1 (en) | Adaptive user interfaces | |
JP5078910B2 (en) | Estimating advertising quality from observed user behavior | |
CN110245301A (en) | A kind of recommended method, device and storage medium | |
US9355095B2 (en) | Click noise characterization model | |
CN109285075A (en) | A kind of Claims Resolution methods of risk assessment, device and server | |
CN113688167A (en) | Deep interest capture model construction method and device based on deep interest network | |
CN111126495B (en) | Model training method, information prediction device, storage medium and equipment | |
TW201939400A (en) | Method and device for determining group of target users | |
Humphreys et al. | Consumer behaviour in lottery: The double hurdle approach and zeros in gambling survey data | |
CN108322317A (en) | A kind of account identification correlating method and server | |
CN110362728A (en) | Information-pushing method, device, equipment and storage medium based on big data analysis | |
CN107368499B (en) | Client label modeling and recommending method and device | |
Kraak et al. | The effect of management choices on the sustainability and economic performance of a mixed fishery: a simulation study | |
Holden et al. | Human judgment vs. quantitative models for the management of ecological resources | |
CN115221396A (en) | Information recommendation method and device based on artificial intelligence and electronic equipment | |
Wang et al. | Sequential evaluation and generation framework for combinatorial recommender system | |
CN110069686A (en) | User behavior analysis method, apparatus, computer installation and storage medium | |
US11458397B1 (en) | Automated real-time engagement in an interactive environment | |
CN113780415B (en) | User portrait generating method, device, equipment and medium based on applet game | |
CN113468394A (en) | Data processing method and device, electronic equipment and storage medium | |
CN107168967B (en) | Target knowledge point acquisition method and device | |
CN112133420A (en) | Data processing method, device and computer readable storage 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 |