CN107368519A - A kind of cooperative processing method and system for agreeing with user interest change - Google Patents
A kind of cooperative processing method and system for agreeing with user interest change Download PDFInfo
- Publication number
- CN107368519A CN107368519A CN201710414491.8A CN201710414491A CN107368519A CN 107368519 A CN107368519 A CN 107368519A CN 201710414491 A CN201710414491 A CN 201710414491A CN 107368519 A CN107368519 A CN 107368519A
- Authority
- CN
- China
- Prior art keywords
- user
- project
- scoring
- targeted customer
- destination item
- 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.)
- Pending
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
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of cooperative processing method for agreeing with user interest change and system, wherein method to include:Gather user data, and build the explicit theme vector of project, obtain the neighbor lists of destination item, to building user interest theme vector, the time of the act information updating user of project is scored project according to user, obtain the neighbor lists of targeted customer, by the neighbor lists of targeted customer and the neighbor lists of destination item, obtain the bulleted list of targeted customer and its neighbour user to destination item, mean value calculation is carried out to the bulleted list of destination item to targeted customer and its neighbour user, obtain prediction scoring of the targeted customer to destination item, using the prediction scoring as recommendation results;The present invention considers influence of the timing to its interest of user behavior, solves the problems, such as concept drift present in prior art, and based on the user interest vector sum user portrait after renewal, score in predicting is carried out to project, has to data preferably anti-openness.
Description
Technical field
The present invention relates to field of information processing, and in particular to it is a kind of agree with user interest change cooperative processing method and
System.
Background technology
With the rapid development of computer network, present we are in the world of an information data explosive growth
Network has the data of flood tide to swap and share daily.Such as online content service, instant messaging, social networks and cloud
The emerging network information services such as calculating play extremely important role in the daily life of today.In face of a feast for the eyes
Application caused by magnanimity information, user, which needs to put into substantial amounts of energy and time according to self-demand, to filter and is screening this
Among the information data that a little network applications provide.
Over time, it is related to be increasingly becoming machine learning, data mining and man-machine interface at present for commending system
The popular research direction in field, GroupLens team, Univ Michigan-Ann Arbor USA Paul such as Univ Minnesota-Twin Cities USA
Many research teams such as team that professor Resnick leads, Microsoft Research have all been carried out much to commending system and its algorithm
Basic research and improvement.Present commending system has obtained significant progress, and substantial amounts of research is solving proposed algorithm presence
Various problems on achieve many achievements.But research be concentrated mainly on solve proposed algorithm existing for scalability problem,
System cold start-up problem etc..With the change and the innovation of every technology increasingly that network world is earth-shaking, the need of user
Also more and more higher is sought, how to solve shortcoming of the current commending system for personalized recommendation ability, is the weight of current industry concern
One of point.Therefore, the discovery of the recessive hobby of user present in personalized recommendation system, user interest drift, recommend when
Effect property is all the Main way studied now.
In personalized recommendation system, due to the diversity and dynamic of user interest, cause be difficult in interaction
The subjective interest of user is specified, i.e., there is the potentiality of user interest.And the collaborative filtering of current relevant user interests change
The user's history behavioral data that algorithm is often directly based simply in data set is similar between user to be calculated
Degree.Thus the problem of many causes commending system accuracy is brought, such as:By matrix represent data set there is
The openness Sparse sex chromosome mosaicism brought, recommendation can not reflect that user interest dynamically changes the user interest drift brought
Problem etc..
The content of the invention
The technical problems to be solved by the invention are in view of the shortcomings of the prior art, there is provided one kind agrees with user interest change
The cooperative processing method and system of change.
The technical scheme that the present invention solves above-mentioned technical problem is as follows:A kind of collaboration processing for agreeing with user interest change
Method, comprise the following steps:
Collection user data simultaneously formats, and obtains user behavior data collection, the user behavior data collection includes user couple
The time of the act information that project scores with user to project;
The explicit theme vector of project is built according to the user behavior data collection, and to the explicit theme vector of the project
Similarity Measure is carried out, obtains the neighbor lists of destination item;
According to Chinese mugwort this great forgetting curve method of guest and the user to user described in the time of the act information updating of project
Project is scored, new user is obtained and project is scored;
According to the user behavior data collection build user interest theme vector, and according to the user interest theme to
Structure user portrait personas is measured, and updates user's portrait personas to project scoring according to the new user;
Similarity Measure is carried out according to user's portrait personas after renewal, obtains the neighbor lists of targeted customer;
According to the neighbor lists of the targeted customer and the neighbor lists of destination item, targeted customer and its neighbour are obtained
Bulleted list of the user to destination item;And
Mean value calculation is carried out to the bulleted list of destination item to the targeted customer and its neighbour user, obtains mesh
Prediction scoring of the user to destination item is marked, using the prediction scoring as recommendation results.
The beneficial effects of the invention are as follows:By user to scoring of the time of the act information updating user of project to project,
Influence of the timing to its interest of user behavior is considered, solves the problems, such as concept drift present in prior art, separately
Outside, based on the user interest vector sum user portrait after renewal, score in predicting is carried out to project, had to data preferably anti-
It is openness.
On the basis of above-mentioned technical proposal, the present invention can also do following improvement.
Further, it is described obtain targeted customer and its neighbour user to the bulleted list of destination item after, in addition to step
Suddenly:
Sequence algorithm is reset according to Bayes to enter the bulleted list of destination item the targeted customer and its neighbour user
Row rearrangement.
It is using the above-mentioned further beneficial effect of scheme:Reordered using Bayes, enhance the overall situation
Stability, it is relevant knowledge in the process reasonable utilization probability theory to reorder that Bayes, which reorders, list and the text of reordering
Sequence consistency constraint between this search listing is considered as both likelihoods, the sequence one of the similar sample of subject information
The constraint of cause property is considered as condition priori, and its final purpose is maximization condition priori and the product of likelihood.
Further, after the prediction scoring for obtaining targeted customer to destination item, in addition to step:
The prediction scoring is screened according to Top-K algorithms, so as to which garbled prediction scoring be tied as recommendation
Fruit.
It is using the above-mentioned further beneficial effect of scheme:Top-K algorithms can filter out the preferred knot being arranged in front
Fruit, improve the reliability of recommendation results.
Further, described to obtain the bulleted list of targeted customer and its neighbour user to destination item, specific method is:
Project is scored according to user new in the neighbor lists of the targeted customer neighbour of the destination item is arranged
Table is ranked up, and obtains the bulleted list of targeted customer and its neighbour user to destination item.
Further, the method for the neighbor lists for obtaining destination item specifically includes:
The explicit theme vector of project is built according to formula (1):
Wherein, MovieTopic (n | i) is the explicit theme vector of project, Rate (Mn) it is project MnIn historic user behavior
The overall score number that data set obtains, TiFor i-th of item subject type dimension;
And the explicit theme vector of the project is calculated according to cosine similarity computational methods, obtain destination item
Neighbor lists:Calculated by cosine similarity formula (2)
Wherein, RateCount (Mn∩Mm) represent project MnWith project MmThe total degree in same user's scoring is appeared in,
RateCount(Mn) represent project MnThe total degree of user's scoring, RateCount (M are concentrated by user behavior datam) represent item
Mesh MmThe total degree of user's scoring is concentrated by user behavior data.
It is using the above-mentioned further beneficial effect of scheme:The explicit theme vector of structure project, can more accurately be expressed
The theme of each project, while mitigate project data because influence of the Deta sparseness in commending system;Consider data
Openness and project hot topic degree is drawn the neighbor lists of project by Similarity Measure, uses cosine on influence caused by project
Similarity calculating method, the neighbor lists of each project are established, objective and accurate project neighbor lists are established for each project.
Further, the method that the renewal user scores project specifically includes:
The time of the act information of project is updated according to Chinese mugwort this great forgetting curve method of guest and the user:It is logical
Cross great this forgetting curve formula (3) the renewal user of Chinese mugwort guest to score to project, obtain new user to project scoring Rateu,n×F
(time),
Wherein, Δ time=(timeu,n-timeu,last) × k, F (time) represent the function of time, wherein timeu,nFor with
Family UuTo project MnThe time scored, estuThe time scored earliest for the user, form are unix timestamps,
hl0For initial half-life period, initial interest attenuation cycle of the user to movie themes is represented, Δ time is half-life period increment coefficient,
timeu,lastFor the user UuTo MnThe intermediate item earliest scoring time, form are unix timestamps, and k is factor of influence,
Rateu,nFor user UuTo project MnScoring.
It is using the above-mentioned further beneficial effect of scheme:The forgetting process of people is simulated, considers user interest by then
Between characteristic global impact, user interest can produce migration etc. change, by this great forgetting curve method of the guest that ends and user to item
Purpose time of the act information is scored to update, and result can be made to have more ageing and accuracy.
Further, the method for the structure user portrait personas specifically includes:
Concentrated in the user behavior data, if targeted customer Uu, user UuInitial interest distribution vector
UserTopicu, useRepresent user interest theme vector, wherein tu,iRepresent user
UuWeighted value in i-th of interest topic type;
And user portrait personas is portrayed according to formula (4),
Wherein, Countu,rateFor user UuThe overall score number to project is concentrated in historic user behavioral data,
Rateu,nFor user UuTo project MnScoring, MovieTopicnFor the explicit theme vector of project, TopicNumnFor project MnTool
Some item subject numbers, Rateu,n× F (time) is that new user scores project, and according to the new user to project
Scoring portrays user's portrait personas to update.
It is using the above-mentioned further beneficial effect of scheme:Consider the different types of all items of user's scoring
Count with the project at itself with the decentralization on type of theme, the objective and accurate consideration user interest of an energy can be depicted
User draws a portrait.
Further, it is described that mean value calculation is carried out to the bulleted list of destination item to targeted customer and its neighbour user
Method specifically include:
Average value is calculated according to formula (5),
Wherein, AverageRateuFor targeted customer UuHistory grade average, Sim (u, v) is targeted customer UuWith
Neighbour user UvSimilarity, Ratev,nFor neighbour user UvTo destination item MnScore value, AverageRatevFor neighbour
User UvScoring average.
It is using the above-mentioned further beneficial effect of scheme:Targeted customer and its neighbour user are calculated to destination item
The average mark of bulleted list has eliminated the influence of user's scoring user's prejudice that may be present as reference value.
Another technical scheme that the present invention solves above-mentioned technical problem is as follows:A kind of collaboration for agreeing with user interest change
Processing system, including:
Acquisition module, for gathering user data and formatting, obtain user behavior data collection, the user behavior number
Include that user scores project and user is to the time of the act information of project according to collection;
First item list builder module, for building the explicit theme vector of project according to the user behavior data collection,
And Similarity Measure is carried out to the explicit theme vector of the project, obtain the neighbor lists of destination item;
Score update module, during for according to Chinese mugwort behavior to project of guest great this forgetting curve method and the user
Between user described in information updating to project score, obtain new user to project score;
User, which draws a portrait, builds module, for building user interest theme vector, and root according to the user behavior data collection
According to user interest theme vector structure user portrait personas, and project is scored according to the new user and updated
User's portrait personas;
User list builds module, for carrying out Similarity Measure according to user's portrait personas after renewal, obtains
The neighbor lists of targeted customer;
Second item list builder module, for the neighbor lists according to the targeted customer and the neighbour of destination item
List, obtain the bulleted list of targeted customer and its neighbour user to destination item;
Score value computing module, for being carried out to the targeted customer and its neighbour user to the bulleted list of destination item
Mean value calculation, prediction scoring of the targeted customer to destination item is obtained, using the prediction scoring as recommendation results.
On the basis of above-mentioned technical proposal, the present invention can also do following improvement.
Further, in addition to screening module, the screening module be used to scoring to the prediction according to Top-K algorithms into
Row screening, so as to using garbled prediction scoring as recommendation results.
Brief description of the drawings
Fig. 1 is the method flow diagram of the cooperative processing method provided in an embodiment of the present invention for agreeing with user interest change;
Fig. 2 is the module frame chart of the coprocessing system provided in an embodiment of the present invention for agreeing with user interest change;
Fig. 3 is the MAE comparing result figures that film provided in an embodiment of the present invention recommends method;
Fig. 4 is the RSME comparing result figures that film provided in an embodiment of the present invention recommends method.
Embodiment
The principle and feature of the present invention are described below in conjunction with accompanying drawing, the given examples are served only to explain the present invention, and
It is non-to be used to limit the scope of the present invention.
With the rapid development of computer network, present we are in the world of an information data explosive growth
Network has the data of flood tide to swap and share daily.Such as online content service, instant messaging, social networks and cloud
The emerging network information services such as calculating play extremely important role in the daily life of today.In face of a feast for the eyes
Application caused by magnanimity information, user, which needs to put into substantial amounts of energy and time according to self-demand, to filter and is screening this
Among the information data that a little network applications provide.
As one of most ripe extensive proposed algorithm of current application, Collaborative Filtering Recommendation Algorithm can be divided mainly into two classes:
Collaborative filtering based on model and the collaborative filtering based on internal memory.It is logical that collaborative filtering based on model essentially consists in consideration
Cross in data set largely as user to data caused by history of project behavior, design available model.It is then based on establishing
To model the future behaviour of targeted customer is predicted, thus produce recommendation results.Now be widely used and study compared with into
Ripe model mainly have the model based on cluster analysis, based on the analysis model of probability topic, the model based on figure, based on square
The model that battle array is decomposed and the model based on naive Bayesian etc..By the difference for the destination object chosen, the collaboration based on internal memory
Filter algorithm can mainly be defined as two classes at present:Based on user's and project-based collaborative filtering.The realization of algorithm is main
It is by the processing to user's score data collection, the similarity between object is calculated, then algorithm for design utilizes what is obtained
The scoring of Similarity Measure prediction user or the preference to project, are finally based on some score in predicting methods and are calculated
To the recommendation results of targeted customer.
For resulting problem of information overload, commending system arises at the historic moment, and commending system can face user huge
Amount information is screened, and is embodied as user and is recommended possible information that is interested, meeting user's request, effectively reduces user and seek
Seek the cost of information needed.
Fig. 1 is the method flow diagram for the cooperative processing method that the present invention agrees with user interest change;
As shown in figure 1, a kind of cooperative processing method for agreeing with user interest change, comprises the following steps:
Step S1:User data is gathered from internet and is formatted, obtains user behavior data collection, user's row
Include user to project scoring, user to the time of the act information of project, the unique ID of user and the unique ID of project for data set
Deng;And preserve into database;
Step S2:The explicit theme vector of project is built according to the user behavior data collection, and the project is explicitly led
Topic vector carries out Similarity Measure, obtains the neighbor lists of destination item;
Step S3:According to the time of the act information updating of Chinese mugwort this great forgetting curve method of guest and the user to project
The user scores project, obtains new user and project is scored;
Step S4:User interest theme vector is built according to the user behavior data collection, and according to the user interest
Theme vector structure user portrait personas, and the user is updated to project scoring according to the new user and drawn a portrait
personas;
Step S5:Similarity Measure is carried out according to user's portrait personas after renewal, obtains the neighbour of targeted customer
List;
Step S6:According to the neighbor lists of the targeted customer and the neighbor lists of destination item, obtain targeted customer and
Bulleted list of its neighbour user to destination item;
Step S7:Item of the sequence algorithm to the targeted customer and its neighbour user to destination item is reset according to Bayes
Mesh list is resequenced;
Step S8:Targeted customer and its neighbour user after reorder are averaged to the bulleted list of destination item
Value calculates, and prediction scoring of the targeted customer to destination item is obtained, using the prediction scoring as recommendation results;
Step S9:The prediction scoring is screened according to Top-K algorithms, so as to which garbled prediction scoring be made
For recommendation results.
In above-described embodiment, by user to scoring of the time of the act information updating user of project to project, it is contemplated that
Influence of the timing of user behavior to its interest, solves the problems, such as concept drift present in prior art, in addition, being based on
User interest vector sum user portrait after renewal, score in predicting is carried out to project, is had to data preferably anti-openness;
Reordered using Bayes, enhance global stability, it is reasonable in the process to reorder that Bayes, which reorders,
With the relevant knowledge of probability theory, the sequence consistency constraint to reorder between list and text search list is considered as both
Likelihood, the sequence consistency constraint of the similar sample of subject information is considered as condition priori, its final purpose is maximum
The product of change condition priori and likelihood;Top-K algorithms can filter out the preferred result being arranged in front, and improve recommendation results
Reliability.
It is traditional when step S7 considers the average mark that calculating neighbour user and targeted customer score in above-described embodiment
Collaborative filtering is to bring user's prejudice sex chromosome mosaicism using the scoring set of whole user as reference, such average mark,
The thought that Bayes is reordered in invention is fused in improved method, neighbour's bulleted list of thus obtained destination item, meter
User is calculated to the average mark of neighbour's Item Sets of destination item as reference value, to eliminate user's scoring user that may be present
The influence of prejudice.Specific practice is to be reordered using Bayes and initial project trust list is reordered, and enhancing is whole
The individual global stability for trusting subgroup, Bayes, which reorders, to be known in the correlation of the process reasonable utilization probability theory to reorder
Know, the sequence consistency constraint to reorder between list and text search list is considered as both likelihoods, subject information
The sequence consistency constraint of similar sample is considered as condition priori, and its final purpose is maximization condition priori and likelihood
Product, eliminate the problem of yardstick that may be present differs.
In above-described embodiment, trust list reorders effect in order to better improve in step S7, and the present invention uses formula
(701) operation is normalized in initial trust list of the method for destination item, and using the Gaussian kernel of formula (702)
Function calculates the similarity weight between two projects, wherein Ar (Ii) represent project IiAll users average score,
wij=exp (- | | Ar (Ii)-Ar(Ij)||) (702)。
Alternatively, it is described to obtain targeted customer on the basis of above-described embodiment as one embodiment of the present of invention
And its neighbour user is to the bulleted list of destination item, specific method:
Project is scored according to user new in the neighbor lists of the targeted customer neighbour of the destination item is arranged
Table is ranked up, and obtains the bulleted list of targeted customer and its neighbour user to destination item.
Alternatively, it is described to obtain destination item on the basis of above-described embodiment as one embodiment of the present of invention
The methods of neighbor lists specifically include:
The explicit theme vector of project is built according to formula (1), can accurately express the theme of each project, simultaneously
Mitigate project data because influence of the Deta sparseness in commending system,
Wherein, MovieTopic (n | i) is the explicit theme vector of project, Rate (Mn) it is project MnIn historic user behavior
The overall score number that data set obtains, TiFor i-th of item subject type dimension;
And the explicit theme vector of the project is calculated according to cosine similarity computational methods, obtain destination item
Neighbor lists:Calculated by cosine similarity formula (2)
Wherein, RateCount (Mn∩Mm) represent project MnWith project MmThe total degree in same user's scoring is appeared in,
RateCount(Mn) represent project MnThe total degree of user's scoring, RateCount (M are concentrated by user behavior datam) represent item
Mesh MmThe total degree of user's scoring is concentrated by user behavior data;The above method, considers Deta sparseness and project is popular
For degree on influence caused by project, proposition project similar value comes improved project similar neighborhoods subgroup, and to the similar near of destination item
Adjacent subgroup reset based on Bayes the rearrangement of sequence algorithm, and more objective and accurate project is finally established for each project
Neighbor lists.
In above-described embodiment, the explicit theme vector of project is built, can more accurately express the theme of each project, together
When mitigate project data because influence of the Deta sparseness in commending system;Consider Deta sparseness and project is popular
Degree draws the neighbor lists of project by Similarity Measure to influence caused by project, using cosine similarity computational methods,
The neighbor lists of each project are established, objective and accurate project neighbor lists are established for each project.
Alternatively, as one embodiment of the present of invention, on the basis of any of the above-described embodiment, described in the renewal
The method that user scores project specifically includes:
The time of the act information of project is updated according to Chinese mugwort this great forgetting curve method of guest and the user:It is logical
Cross great this forgetting curve formula (3) the renewal user of Chinese mugwort guest to score to project, obtain new user to project scoring Rateu,n×F
(time),
Wherein, Δ time=(timeu,n-timeu,last) × k, F (time) represent the function of time, wherein timeu,nFor with
Family UuTo project MnThe time scored, estuThe time scored earliest for the user, form are unix timestamps,
hl0For initial half-life period, initial interest attenuation cycle of the user to movie themes is represented, Δ time is half-life period increment coefficient,
timeu,lastFor the user UuTo MnThe intermediate item earliest scoring time, form are unix timestamps, and k is factor of influence,
Rateu,nFor user UuTo project MnScoring.
The renewal scored above project user, its Specific Principles are:
(1) interest more than all occurred in long-time in the historical behavior of user or occurrence number, finally using
It can occupy larger weights in the interest distribution of family.On the contrary, the interest that time of occurrence is shorter or occurrence number is few, shared power
Value can be smaller, and this regulation mechanism is that the behavior record based on user is carried out automatically;
(2) characteristic of sequential is combined, transition over time, is never occurred in follow-up user behavior sequential
Interest, its weighing factor can gradually reduce and remain to a low level, and this meets the forgetting law of people;
(3) ensure that the weights of each interest in the interest distribution vector of end user add and about 1.
In above-described embodiment, the forgetting process of people is simulated, considers global impact of the user interest by time response,
User interest can produce the change such as migration, the time of the act information by this great forgetting curve method of the guest that ends and user to project
To update scoring, result can be made to have more ageing and accuracy.
Alternatively, as one embodiment of the present of invention, on the basis of above-described embodiment, the structure user portrait
Personas method specifically includes:
Concentrated in the user behavior data, if targeted customer Uu, user UuInitial interest distribution vector
UserTopicu, useRepresent user interest theme vector, wherein tu,iRepresent user Uu
Weighted value in i-th of interest topic type;
And user portrait personas is portrayed according to formula (4),
Wherein, Countu,rateFor user UuThe overall score number to project is concentrated in historic user behavioral data,
Rateu,nFor user UuTo project MnScoring, MovieTopicnFor the explicit theme vector of project, TopicNumnFor project MnTool
Some item subject numbers, such as a film have 3 theme dimensional attributes, that TopicNumnAs 3, and so on,
Rateu,n× F (time) is that new user scores project, and portrays use to project scoring to update according to the new user
Draw a portrait personas at family.
Above-described embodiment is it is to be understood that Personas are a concrete representation of target
Users. " Persona is the virtual representations of real user, be built upon a series of True Datas (Marketing data,
Usability data) on targeted customer's model, by user investigation go understand user, according to their target, behavior
With the difference of viewpoint, they are divided into different types, characteristic feature is then extracted in each type, name is assigned, shines
Piece, some demography key element, scenes etc. describe, and are formed a user archetype (personas).
Specifically, considering that user's portrait of user interest decay is portrayed, implementation process is:
Input:Targeted customer Uu, user UuInitial interest distribution vector UserTopicu;
Output:User UuRenewal after new interest distribution vector UserTopicu *;
Step is:
01:Initialize UserTopicu *;
02:From targeted customer UuHistorical behavior extracting data Countu,rate, timeu,n, estu, timeu,last;
03:According to targeted customer UuHistorical behavior data in obtain the user have carry out historical behavior bulleted list,
And extract the TopicNum of respective itemnThe obtained explicit theme vector MovieTopic of project is handled with through step 1n;
04:Substitute into formula
Complete targeted customer UuBased on project MnAgree with the scoring renewal of user interest decay;(circulation 1)
05:Substitute into formula
Complete user UuUser portrait renewal;(circulation 2)
06:End loop 1;
07:End loop 2.
In above-described embodiment, consider that the different types of all items number of user's scoring and the project are same at itself
Decentralization on type of theme, the objective and accurate user's portrait for considering user interest of an energy can be depicted.
Alternatively, as one embodiment of the present of invention, it is described to targeted customer and its neighbour user to destination item
The method that bulleted list carries out mean value calculation specifically includes:
Average value is calculated according to formula (5),
Wherein, AverageRateuFor targeted customer UuHistory grade average, Sim (u, v) is targeted customer UuWith
Neighbour user UvSimilarity, Ratev,nFor neighbour user UvTo destination item MnScore value, AverageRatevFor neighbour
User UvScoring average.
In above-described embodiment, calculate targeted customer and its neighbour user is allocated as to being averaged for bulleted list of destination item
For reference value, the influence of user's scoring user's prejudice that may be present has been eliminated.
Fig. 2 is the module frame chart of the coprocessing system provided in an embodiment of the present invention for agreeing with user interest change;
Alternatively, as an alternative embodiment of the invention, as shown in Fig. 2 a kind of collaboration for agreeing with user interest change
Processing system, including:
Acquisition module, for gathering user data and formatting, obtain user behavior data collection, the user behavior number
Include that user scores project and user is to the time of the act information of project according to collection;
First item list builder module, for building the explicit theme vector of project according to the user behavior data collection,
And Similarity Measure is carried out to the explicit theme vector of the project, obtain the neighbor lists of destination item;
Score update module, during for according to Chinese mugwort behavior to project of guest great this forgetting curve method and the user
Between user described in information updating to project score, obtain new user to project score;
User, which draws a portrait, builds module, for building user interest theme vector, and root according to the user behavior data collection
According to user interest theme vector structure user portrait personas, and project is scored according to the new user and updated
User's portrait personas;
User list builds module, for carrying out Similarity Measure according to user's portrait personas after renewal, obtains
The neighbor lists of targeted customer;
Second item list builder module, for the neighbor lists according to the targeted customer and the neighbour of destination item
List, obtain the bulleted list of targeted customer and its neighbour user to destination item;
Score value computing module, for being carried out to the targeted customer and its neighbour user to the bulleted list of destination item
Mean value calculation, prediction scoring of the targeted customer to destination item is obtained, using the prediction scoring as recommendation results.
Alternatively, it is used for basis as one embodiment of the present of invention, in addition to screening module, the screening module
Top-K algorithms screen to the prediction scoring, so as to using garbled prediction scoring as recommendation results.
Fig. 3 is the MAE comparing result figures that film provided in an embodiment of the present invention recommends method;
Fig. 4 is the RSME comparing result figures that film provided in an embodiment of the present invention recommends method;
A specific embodiment is named to illustrate:
Based on the cooperative processing method for agreeing with user interest change, the film that user's scoring is provided in this instantiation pushes away
Method is recommended, as shown in Figure 3-4,
Data set contains 943 users to 100000 of 1682 films scoring records and corresponding time of the act,
And each user's at least 20 behavior records, its score 1-5/.The rating matrix density of user and film is:
Parameter prediction:
For simplicity in this example, α and β are used as default value, wherein iteration time using α=50/K and β=0.01
Number is defaulted as L=500.Describe as shown in Figure 4 and data set is being divided into 8:In the case of 2 training set and test set, balance
Influence of the factor lambda to recommendation accuracy rate, wherein accuracy rate (Precision) are defined as in Top-N recommendation lists, really
Correctly made the ratio shared by the project of recommendation.Shown in calculation formula such as formula (101):
System integrally recommends accuracy rate to present first to increase with balance factor λ increase and subtract afterwards, is finally intended to stabilization
Trend.And when λ is in the low order of magnitude and high quantity level position, the change of accuracy rate is little, main reason is that this reality
The λ of order of magnitude change asymptotic is intended to stabilization, and λ taking between 0-1 for the degree of influence of user interest in example
Value, system can show preferably to recommend performance on the whole.This example selects λ=0.2.
Specific embodiment method timing impact evaluation:
For the importance in the inventive method sequential stage, this specific embodiment uses following appraisal procedure:
(1) user U is extracted according to original stepsiMaximum sequential sequence of interest be
(2) it is right in the step in second stage interest personalization more new algorithmIn each interest Tj, using inverse
To extraction.If primal algorithm is that chronologically sequence of interest from left to right extracts interest successively, and make the calculating of two layers of circulation,
And now then extract interest successively from right to left and make two layers of cycle calculations of identical.
On the basis of other steps and original method identical, the inventive method and above method method (are defined as
Time-reverse Top-5) is carried out in the case of potential interest dimension K differences and recommends Experimental comparison.In view of Gibbs
The randomness of Sampling processes, experiment is related to ten assessment data altogether, and takes its average value as end product.
As seen from the figure, the way of recommendation of positive sequential is used in the inventive method really on system recommendation performance is improved
Make great sense, this is also a kind of important performance of the present invention on solving the problems, such as concept drift.I.e. in the time
During transition, the interest of user can change with some implicit scenes, only by holding and analyzing user behavior
Temporal characteristicses, and caused system change can just effectively improve recommendation effect.
Example recommends performance comprehensive assessment:
As shown in figure 3, respectively describing in the case of different proportion training set, this specific embodiment is recommending performance
The recommendation effect figure of upper accuracy rate, recall rate and F values, wherein recall rate (Recall) are then defined as in whole test sets
In, the ratio shared by the project of recommendation is correctly made, shown in calculation formula such as formula (102):
Because in some cases, accuracy rate and recall rate can only reflect single aspect, there is the characteristic of contradiction, i.e.,
One high and one low situation.Therefore the measure of F values is used, the two is integrated.Shown in calculation formula such as formula (103):
The anti-openness assessment of example score in predicting:
As shown in figure 4, it is that the inventive method is combined the collaborative recommendation method to be formed with collaborative filtering in this example
And collaborative filtering (UserBased_CF) method based on user using Pearson came similarity measurement (Proposed_CF)
Comparing result.As seen from the figure, in the case of Sparse, new collaborative recommendation method (Proposed_CF) can obtain smaller
Mean square error.It is main reason is that this method is that the potential interest based on user carries out similarity measurement, the mistake indirectly
Journey can reduce the influence that Deta sparseness in collaborative filtering extracts the stage to neighbour user effectively in low dimensional.Wherein
Error larger to data RMSE is more sensitive.The recommendation effect that its value is smaller to represent algorithm is better.If the user of prediction comments
Diversity is { predict1,predict2,…,predictn, corresponding true scoring is { real1,real2,…,realn, then
The method and system of the present invention by user to scoring of the time of the act information updating user of project to project,
Influence of the timing to its interest of user behavior is considered, solves the problems, such as concept drift present in prior art, separately
Outside, based on the user interest vector sum user portrait after renewal, score in predicting is carried out to project, had to data preferably anti-
It is openness.
Reader should be understood that in the description of this specification, reference term " one embodiment ", " some embodiments ", " show
The description of example ", " specific example " or " some examples " etc. means to combine the specific features that the embodiment or example describe, tied
Structure, material or feature are contained at least one embodiment or example of the present invention.In this manual, to above-mentioned term
Schematic representation need not be directed to identical embodiment or example.Moreover, description specific features, structure, material or
Feature can combine in an appropriate manner in any one or more embodiments or example.In addition, in not conflicting situation
Under, those skilled in the art by the different embodiments or example described in this specification and different embodiments or can show
The feature of example is combined and combined.
It is apparent to those skilled in the art that for convenience of description and succinctly, the dress of foregoing description
The specific work process with unit is put, the corresponding process in preceding method embodiment is may be referred to, will not be repeated here.
In several embodiments provided herein, it should be understood that disclosed apparatus and method, can pass through
Other modes are realized.For example, device embodiment described above is only schematical, for example, the division of unit, only
Only a kind of division of logic function, there can be other dividing mode when actually realizing, such as multiple units or component can be tied
Another system is closed or is desirably integrated into, or some features can be ignored, or do not perform.
The unit illustrated as separating component can be or may not be physically separate, be shown as unit
Part can be or may not be physical location, you can with positioned at a place, or multiple nets can also be distributed to
On network unit.Some or all of unit therein can be selected to realize scheme of the embodiment of the present invention according to the actual needs
Purpose.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, can also
It is that unit is individually physically present or two or more units are integrated in a unit.It is above-mentioned integrated
Unit can both be realized in the form of hardware, can also be realized in the form of SFU software functional unit.
If integrated unit realized in the form of SFU software functional unit and as independent production marketing or in use,
It can be stored in a computer read/write memory medium.Based on such understanding, technical scheme substantially or
Person says the part to be contributed to prior art, or all or part of the technical scheme can be in the form of software product
Embody, the computer software product is stored in a storage medium, including some instructions are causing a computer
Equipment (can be personal computer, server, or network equipment etc.) perform each embodiment method of the present invention whole or
Part steps.And foregoing storage medium includes:USB flash disk, mobile hard disk, read-only storage (ROM, Read-Only Memory),
Random access memory (RAM, Random Access Memory), magnetic disc or CD etc. are various can be with store program codes
Medium.
More than, it is only embodiment of the invention, but protection scope of the present invention is not limited thereto, and it is any ripe
Know those skilled in the art the invention discloses technical scope in, various equivalent modifications can be readily occurred in or replaced
Change, these modifications or substitutions should be all included within the scope of the present invention.Therefore, protection scope of the present invention Ying Yiquan
The protection domain that profit requires is defined.
Claims (10)
1. a kind of cooperative processing method for agreeing with user interest change, it is characterised in that comprise the following steps:
Collection user data simultaneously formats, and obtains user behavior data collection, and the user behavior data collection includes user to project
The time of the act information of scoring and user to project;
The explicit theme vector of project is built according to the user behavior data collection, and phase is carried out to the explicit theme vector of the project
Calculated like degree, obtain the neighbor lists of destination item;
According to Chinese mugwort this great forgetting curve method of guest and the user to user described in the time of the act information updating of project to item
Mesh scores, and obtains new user and project is scored;
User interest theme vector is built according to the user behavior data collection, and built according to the user interest theme vector
User portrait personas, and user's portrait personas is updated to project scoring according to the new user;
Similarity Measure is carried out according to user's portrait personas after renewal, obtains the neighbor lists of targeted customer;
According to the neighbor lists of the targeted customer and the neighbor lists of destination item, targeted customer and its neighbour user couple are obtained
The bulleted list of destination item;And
Mean value calculation is carried out to the bulleted list of destination item to the targeted customer and its neighbour user, obtains targeted customer
Prediction scoring to destination item, using the prediction scoring as recommendation results.
2. the cooperative processing method according to claim 1 for agreeing with user interest change, it is characterised in that described to obtain mesh
After user and its neighbour user are marked to the bulleted list of destination item, in addition to step:
Sequence algorithm is reset according to Bayes and carries out weight to the bulleted list of destination item to the targeted customer and its neighbour user
New sort.
3. the cooperative processing method according to claim 1 for agreeing with user interest change, it is characterised in that described to obtain mesh
After marking prediction scoring of the user to destination item, in addition to step:
The prediction scoring is screened according to Top-K algorithms, so as to using garbled prediction scoring as recommendation results.
4. the cooperative processing method according to claim 1 for agreeing with user interest change, it is characterised in that described to obtain mesh
The bulleted list of user and its neighbour user to destination item is marked, specific method is:
Project is scored to enter the neighbor lists of the destination item according to user new in the neighbor lists of the targeted customer
Row sequence, obtains the bulleted list of targeted customer and its neighbour user to destination item.
5. the cooperative processing method according to claim 1 for agreeing with user interest change, it is characterised in that described to obtain mesh
The method of the neighbor lists of mark project specifically includes:
The explicit theme vector of project is built according to formula (1):
Wherein, MovieTopic (n | i) is the explicit theme vector of project, Rate (Mn) it is project MnIn historic user behavioral data
Collect the overall score number obtained, TiFor i-th of item subject type dimension;
And the explicit theme vector of the project is calculated according to cosine similarity computational methods, obtain the neighbour of destination item
List:Calculated by cosine similarity formula (2),
Wherein, RateCount (Mn∩Mm) represent project MnWith project MmThe total degree in same user's scoring is appeared in,
RateCount(Mn) represent project MnThe total degree of user's scoring, RateCount (M are concentrated by user behavior datam) represent item
Mesh MmThe total degree of user's scoring is concentrated by user behavior data.
6. the cooperative processing method according to claim 1 for agreeing with user interest change, it is characterised in that the renewal institute
The method that user scores to project is stated to specifically include:
The time of the act information of project is updated according to Chinese mugwort this great forgetting curve method of guest and the user:Pass through the guest that ends
Great this forgetting curve formula (3) renewal user scores project, obtains new user to project scoring Rateu,n× F (time),
Wherein, Δ time=(timeu,n-timeu,last) × k, F (time) represent the function of time, wherein timeu,nFor user UuIt is right
Project MnThe time scored, estuThe time scored earliest for the user, form are unix timestamps, hl0To be first
Begin half-life period, represent initial interest attenuation cycle of the user to movie themes, Δ time is half-life period increment coefficient, timeu,last
For the user UuTo MnThe intermediate item earliest scoring time, form are unix timestamps, and k is factor of influence, Rateu,nFor user
UuTo project MnScoring.
7. the cooperative processing method according to claim 6 for agreeing with user interest change, it is characterised in that the structure is used
Family portrait personas method specifically includes:
Concentrated in the user behavior data, if targeted customer Uu, user UuInitial interest distribution vector UserTopicu, useRepresent user interest theme vector, wherein tu,iRepresent user UuIn i-th of interest master
Inscribe the weighted value in type;
And user portrait personas is portrayed according to formula (4),
Wherein, Countu,rateFor user UuThe overall score number to project, Rate are concentrated in historic user behavioral datau,nFor with
Family UuTo project MnScoring, MovieTopicnFor the explicit theme vector of project, TopicNumnFor project MnThe project master having
Inscribe number, Rateu,n× F (time) is that new user scores project, and project scoring is carved to update according to the new user
Draw user's portrait personas.
8. the cooperative processing method according to claim 1 for agreeing with user interest change, it is characterised in that described to target
The method that user and its neighbour user carry out mean value calculation to the bulleted list of destination item specifically includes:
Average value is calculated according to formula (5),
Wherein, AverageRateuFor targeted customer UuHistory grade average, Sim (u, v) is targeted customer UuUsed with neighbour
Family UvSimilarity, Ratev,nFor neighbour user UvTo destination item MnScore value, be neighbour AverageRatevUser Uv's
Score average.
A kind of 9. coprocessing system for agreeing with user interest change, it is characterised in that including:
Acquisition module, for gathering user data and formatting, obtain user behavior data collection, the user behavior data Ji Bao
User is included to project scoring and user to the time of the act information of project;
First item list builder module, for building the explicit theme vector of project according to the user behavior data collection, and it is right
The explicit theme vector of project carries out Similarity Measure, obtains the neighbor lists of destination item;
Score update module, for according to Chinese mugwort this great forgetting curve method of guest and the user to the time of the act information of project
Update the user to score to project, obtain new user and project is scored;
User, which draws a portrait, builds module, for building user interest theme vector according to the user behavior data collection, and according to institute
User interest theme vector structure user portrait personas is stated, and the use is updated to project scoring according to the new user
Draw a portrait personas at family;
User list builds module, for carrying out Similarity Measure according to user's portrait personas after renewal, obtains target
The neighbor lists of user;
Second item list builder module, for the neighbor lists and the neighbor lists of destination item according to the targeted customer,
Obtain the bulleted list of targeted customer and its neighbour user to destination item;
Score value computing module, for carrying out average value to the bulleted list of destination item to the targeted customer and its neighbour user
Calculate, prediction scoring of the targeted customer to destination item is obtained, using the prediction scoring as recommendation results.
10. the coprocessing system according to claim 9 for agreeing with user interest change, it is characterised in that also include sieve
Modeling block, the screening module is used to screen the prediction scoring according to Top-K algorithms, so as to by garbled prediction
Scoring is used as recommendation results.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710414491.8A CN107368519A (en) | 2017-06-05 | 2017-06-05 | A kind of cooperative processing method and system for agreeing with user interest change |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710414491.8A CN107368519A (en) | 2017-06-05 | 2017-06-05 | A kind of cooperative processing method and system for agreeing with user interest change |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107368519A true CN107368519A (en) | 2017-11-21 |
Family
ID=60305427
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710414491.8A Pending CN107368519A (en) | 2017-06-05 | 2017-06-05 | A kind of cooperative processing method and system for agreeing with user interest change |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107368519A (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108133013A (en) * | 2017-12-22 | 2018-06-08 | 平安养老保险股份有限公司 | Information processing method, device, computer equipment and storage medium |
CN108288179A (en) * | 2018-01-25 | 2018-07-17 | 链家网(北京)科技有限公司 | A kind of user preference source of houses computational methods and system |
CN108287902A (en) * | 2018-01-24 | 2018-07-17 | 厦门大学 | A kind of commending system method based on Data Missing Mechanism |
CN108334575A (en) * | 2018-01-23 | 2018-07-27 | 北京三快在线科技有限公司 | A kind of recommendation results sequence modification method and device, electronic equipment |
CN108595533A (en) * | 2018-04-02 | 2018-09-28 | 深圳大学 | A kind of item recommendation method, storage medium and server based on collaborative filtering |
CN108664558A (en) * | 2018-04-04 | 2018-10-16 | 山东科技大学 | A kind of Web TV personalized ventilation system method towards large-scale consumer |
CN109168044A (en) * | 2018-10-11 | 2019-01-08 | 北京奇艺世纪科技有限公司 | A kind of determination method and device of video features |
CN109255522A (en) * | 2018-08-14 | 2019-01-22 | 殷肇良 | A kind of user management method |
CN109389168A (en) * | 2018-09-29 | 2019-02-26 | 国信优易数据有限公司 | Project recommendation model training method, item recommendation method and device |
CN110008405A (en) * | 2019-03-25 | 2019-07-12 | 华南理工大学 | A kind of personalization message method for pushing and system based on timeliness |
CN110210944A (en) * | 2019-06-05 | 2019-09-06 | 齐鲁工业大学 | The multitask recommended method and system of joint Bayesian inference and weighting refusal sampling |
CN111858702A (en) * | 2020-06-28 | 2020-10-30 | 西安工程大学 | User behavior data acquisition and weighting method for dynamic portrait |
CN112231576A (en) * | 2020-11-05 | 2021-01-15 | 中国联合网络通信集团有限公司 | Target user determination method and device |
CN114025205A (en) * | 2021-11-02 | 2022-02-08 | 天津大学 | Intelligent recommendation method for home TV video |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090171763A1 (en) * | 2007-12-31 | 2009-07-02 | Yahoo! Inc. | System and method for online advertising driven by predicting user interest |
CN103617540A (en) * | 2013-10-17 | 2014-03-05 | 浙江大学 | E-commerce recommendation method of tracking user interest changes |
CN104063481A (en) * | 2014-07-02 | 2014-09-24 | 山东大学 | Film individuation recommendation method based on user real-time interest vectors |
CN106682121A (en) * | 2016-12-09 | 2017-05-17 | 广东工业大学 | Time utility recommendation method based on interest change of user |
-
2017
- 2017-06-05 CN CN201710414491.8A patent/CN107368519A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090171763A1 (en) * | 2007-12-31 | 2009-07-02 | Yahoo! Inc. | System and method for online advertising driven by predicting user interest |
CN103617540A (en) * | 2013-10-17 | 2014-03-05 | 浙江大学 | E-commerce recommendation method of tracking user interest changes |
CN104063481A (en) * | 2014-07-02 | 2014-09-24 | 山东大学 | Film individuation recommendation method based on user real-time interest vectors |
CN106682121A (en) * | 2016-12-09 | 2017-05-17 | 广东工业大学 | Time utility recommendation method based on interest change of user |
Non-Patent Citations (2)
Title |
---|
唐晓波等: "基于主题的用户兴趣模型的构建及动态更新", 《情报理论与实践》 * |
邓星: "基于协同过滤的推荐方法的研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108133013A (en) * | 2017-12-22 | 2018-06-08 | 平安养老保险股份有限公司 | Information processing method, device, computer equipment and storage medium |
CN108133013B (en) * | 2017-12-22 | 2021-02-09 | 平安养老保险股份有限公司 | Information processing method, information processing device, computer equipment and storage medium |
CN108334575A (en) * | 2018-01-23 | 2018-07-27 | 北京三快在线科技有限公司 | A kind of recommendation results sequence modification method and device, electronic equipment |
CN108334575B (en) * | 2018-01-23 | 2022-04-26 | 北京三快在线科技有限公司 | Recommendation result sorting correction method and device and electronic equipment |
CN108287902A (en) * | 2018-01-24 | 2018-07-17 | 厦门大学 | A kind of commending system method based on Data Missing Mechanism |
CN108287902B (en) * | 2018-01-24 | 2020-11-20 | 厦门大学 | Recommendation system method based on data non-random missing mechanism |
CN108288179B (en) * | 2018-01-25 | 2021-02-02 | 贝壳找房(北京)科技有限公司 | User preference house source calculation method and system |
CN108288179A (en) * | 2018-01-25 | 2018-07-17 | 链家网(北京)科技有限公司 | A kind of user preference source of houses computational methods and system |
CN108595533A (en) * | 2018-04-02 | 2018-09-28 | 深圳大学 | A kind of item recommendation method, storage medium and server based on collaborative filtering |
CN108595533B (en) * | 2018-04-02 | 2021-09-14 | 深圳大学 | Article recommendation method based on collaborative filtering, storage medium and server |
CN108664558A (en) * | 2018-04-04 | 2018-10-16 | 山东科技大学 | A kind of Web TV personalized ventilation system method towards large-scale consumer |
CN108664558B (en) * | 2018-04-04 | 2020-05-05 | 山东科技大学 | Network television personalized recommendation service method for large-scale users |
CN109255522A (en) * | 2018-08-14 | 2019-01-22 | 殷肇良 | A kind of user management method |
CN109389168A (en) * | 2018-09-29 | 2019-02-26 | 国信优易数据有限公司 | Project recommendation model training method, item recommendation method and device |
CN109389168B (en) * | 2018-09-29 | 2021-07-13 | 国信优易数据股份有限公司 | Project recommendation model training method, project recommendation method and device |
CN109168044A (en) * | 2018-10-11 | 2019-01-08 | 北京奇艺世纪科技有限公司 | A kind of determination method and device of video features |
CN110008405A (en) * | 2019-03-25 | 2019-07-12 | 华南理工大学 | A kind of personalization message method for pushing and system based on timeliness |
CN110210944B (en) * | 2019-06-05 | 2021-04-23 | 齐鲁工业大学 | Multi-task recommendation method and system combining Bayesian inference and weighted rejection sampling |
CN110210944A (en) * | 2019-06-05 | 2019-09-06 | 齐鲁工业大学 | The multitask recommended method and system of joint Bayesian inference and weighting refusal sampling |
CN111858702A (en) * | 2020-06-28 | 2020-10-30 | 西安工程大学 | User behavior data acquisition and weighting method for dynamic portrait |
CN111858702B (en) * | 2020-06-28 | 2022-02-11 | 西安工程大学 | User behavior data acquisition and weighting method for dynamic portrait |
CN112231576A (en) * | 2020-11-05 | 2021-01-15 | 中国联合网络通信集团有限公司 | Target user determination method and device |
CN112231576B (en) * | 2020-11-05 | 2023-06-06 | 中国联合网络通信集团有限公司 | Target user determination method and device |
CN114025205A (en) * | 2021-11-02 | 2022-02-08 | 天津大学 | Intelligent recommendation method for home TV video |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107368519A (en) | A kind of cooperative processing method and system for agreeing with user interest change | |
McMahan et al. | Ad click prediction: a view from the trenches | |
CN103678518B (en) | Method and device for adjusting recommendation lists | |
CN105320719B (en) | A kind of crowd based on item label and graphics relationship raises website item recommended method | |
CN105760443B (en) | Item recommendation system, project recommendation device and item recommendation method | |
CN108647251A (en) | The recommendation sort method of conjunctive model is recycled based on wide depth door | |
CN104166668A (en) | News recommendation system and method based on FOLFM model | |
CN110427560A (en) | A kind of model training method and relevant apparatus applied to recommender system | |
CN107526810B (en) | Method and device for establishing click rate estimation model and display method and device | |
CN112749330B (en) | Information pushing method, device, computer equipment and storage medium | |
CN112508609B (en) | Crowd expansion prediction method, device, equipment and storage medium | |
CN108491511A (en) | Data digging method and device, model training method based on diagram data and device | |
Wang et al. | Interactive social recommendation | |
CN106610970A (en) | Collaborative filtering-based content recommendation system and method | |
CN106708953A (en) | Discrete particle swarm optimization based local community detection collaborative filtering recommendation method | |
CN115917577A (en) | System and method for next basket recommendation using dynamic attribute modeling | |
CN109460519A (en) | Browse object recommendation method and device, storage medium, server | |
CN104008204B (en) | A kind of dynamic multidimensional context aware film commending system and its implementation | |
CN116304299A (en) | Personalized recommendation method integrating user interest evolution and gradient promotion algorithm | |
CN113326432A (en) | Model optimization method based on decision tree and recommendation method | |
CN113656699B (en) | User feature vector determining method, related equipment and medium | |
CN110472659A (en) | Data processing method, device, computer readable storage medium and computer equipment | |
CN114693409A (en) | Product matching method, device, computer equipment, storage medium and program product | |
Wu et al. | Temporal bipartite projection and link prediction for online social networks | |
CN110851708B (en) | Negative sample extraction method, device, computer equipment and 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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171121 |
|
RJ01 | Rejection of invention patent application after publication |