CN106951547A - A kind of cross-domain recommendation method based on intersection user - Google Patents
A kind of cross-domain recommendation method based on intersection user Download PDFInfo
- Publication number
- CN106951547A CN106951547A CN201710188924.2A CN201710188924A CN106951547A CN 106951547 A CN106951547 A CN 106951547A CN 201710188924 A CN201710188924 A CN 201710188924A CN 106951547 A CN106951547 A CN 106951547A
- Authority
- CN
- China
- Prior art keywords
- user
- domain
- prime
- cross
- intersecting
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Abstract
The invention belongs to Internet technical field, disclose a kind of based on the cross-domain recommendation method for intersecting user, it is described that the data of source domain and aiming field are handled based on the cross-domain recommendation method for intersecting user, filter out intersection user, and data fusion of the user in two domains will be intersected, constitute a new rating matrix;Then fraction homogenization is handled, calculated using user's calculating formula of similarity.The data that the present invention can fully rely on original field calculate user's similarity, final to produce recommendation, and user's cold start-up is solved the problems, such as to a certain extent;Improvement user's calculating formula of similarity that user intersects user is proposed, new cross-domain proposed algorithm is formd;Based on the cross-domain proposed algorithm for intersecting user, migration intersects data of the user in source domain into aiming field, strengthens the data rich degree of aiming field, so as to obtain more accurate personalized recommendation result.
Description
Technical field
The invention belongs to Internet technical field, more particularly to a kind of cross-domain recommendation method based on intersection user.
Background technology
With the development of information technology, the various aspects that internet has been permeated in life, we are in a prosperity and development
Internet era.The information of magnanimity is there is on current internet, daily people can be from mobile phone, computer, TV, intelligent hand
Table, can so that obtain numerous and diverse information on station large-size screen monitors, elevator building screen, and therefrom to select the part useful to oneself
The information of acquisition is too many, but available information interested in other words is very little, and this phenomenon claims just to be called " information overload ".To answer
Pair with solving this problem, major commercial companies and scientific research institution are proposed effective method, such as portal website,
Information classification is enumerated on webpage, user can find the consulting of news, music and stock from a website.From door
The functional localization of website can be seen that its classification is very general, and user can not also obtain the body of personalization from portal website
Test, what everyone saw is identical information.With the development of internet, search engine, which is born and becomes people, obtains letter
The important means of breath.The web crawlers of search engine is shuttled in internet, and summary info is captured to website as much as possible, is used
Retrieved on a search engine using customized keyword at family, it is possible to obtain the information of correlation.Search engine is solving information
It is made that important contribution also becomes most successful internet in the world there is provided the Google of search engine on overload problem public
One of department.But people are not that the demand of oneself can be summarized as into keyword, and for same keyword search engine
May also provide with the completely unrelated content of user's actual requirement, for example, user search " millet ", it is possible to be intended to be closed
In the information of millet mobile phone, the eating method of such as millet, nutritive value, planting technology, it is also possible to be desirable to search apple
The relevant information of company, this can not judge completely for search engine.In order to meet the individual demand of user, people
Again createed personalized recommendation system.Personalized recommendation system is a subset of information filtering system, and it is directed to pre-
Survey preference and evaluation of the user for target item.Propose that system filtering recommendation algorithms are opened from Goldberg in 1992 et al.
Begin, various proposed algorithms are continued to bring out like the mushrooms after rain, and much collaborative filtering is all based on wherein having.At present, cooperateed with
Filter has developed into the system filtering based on history and the major class of collaborative filtering two based on model.Collaborative filtering based on history
By the usage history of digging user, calculate the similarity between data and propose to recommend, the recommendations of GroupLens initially are calculated
Method --- the collaborative filtering based on user is as such, and other algorithms include project-based collaborative filtering and are mixed with item
Collaborative filtering of mesh-user etc.;Collaborative filtering based on model, which then efforts be made so that, build model with historical data study, use
Model is recommended, and the algorithm of this respect includes the various commending systems based on matrix decomposition, such as based on singular value decomposition
SVD, Non-negative Matrix Factorization NMF model construction, also have some to be based on BP neural network, Bayes classifier etc. by model construction
The method that problem is converted into classification.At present, ACM SIGKDD international conferences have many papers that commending system, ACM are discussed every year
RecSyc meetings were all held as scheduled every year since 2007, proposed algorithm, commending system be current academia's data mining with
One important subject in Knowledge Discovery field.Outside academia, personalized recommendation algorithm has had been applied to major business
Among industry commending system, wherein most successful case is exactly commercial product recommending systems of the Amazon (Amazon) from 1999 releases, it
Understand the purchase based on user and browse history, the commodity that personalized displaying user may need.Current domestic major electricity
Business website is also all proposed similar commending system, including the easily purchase etc. of Jingdone district, Taobao and Suning.Except electric business field, video
Class website Netflix, bean cotyledon can recommend the film that may like to user, and technology class website CSDN reads blog articles in user
Other blog articles of related subject can be also recommended to browse reference for user afterwards.Amazon once issued data display its books sales volume
30% is all brought by personalized recommendation system, and this is efficiently and effectively improved from one side explanation personalized recommendation system
Pageview of the user to website.Just because of this, the match of proposed algorithm and commending system was all held in major websites, such as
Netflix has held the Netflix Prize contests of 3 years by a definite date, it is intended to improve the accuracy of the current commending systems of Netflix;
Individual or team for that can improve 10% on the basis of current commending system accuracy, it is beautiful that Netflix will reward 1,000,000
Member.Tengxun patronage in 2012 KDD Cup2012, Baidu is in the 2013 film commending system algorithm contests held and Ali
The big data held in 2014 recommends contest all to embody the domestic attention to commending system algorithm.Personalized recommendation system is passed through
The development for spending decades has reached its maturity, but its some problem existed never has and solved at all, such as cold start-up
Problem, Sparse sex chromosome mosaicism, it is few in system conceptual data amount, or in the case that the data volume that provides of user is few, it is traditional
Personalized recommendation system has no idea to bring high-quality recommended project well for user.The fast development of mobile Internet, enriches
People access the approach of internet, and the widely using of smart mobile phone allows people forever in " online " state, each
The data volume that people produces in various network services is also increasing.If user can be accumulated in other network services
Data are used in target commending system, then can greatly alleviate cold start-up and sparse sex chromosome mosaicism in some cases, this
It is exactly cross-domain commending system.Meanwhile, cross-domain commending system can also recommend the further types of article of user, for example, user exists
If the score data on bean cotyledon may apply on the commending system of electric business website, more accurately can recommend for user can
The film CD that be able to can buy, OST and film original work books etc. derived from film.With the flourishing hair of Internet industry
Exhibition, user is required for the information in face of magnanimity daily, and how therefrom to filter out important information turns into one of internet industry
Major issue.In order to solve problem of information overload, there has been proposed various schemes, wherein most successful, most widely used at present
It is exactly personalized recommendation system.Commending system sums up user interest feature by analyzing user history information, so as to predict use
Family is to the hobby for the article do not checked.Personalized recommendation system passes through the development of more than 20 years, big academicly emerging
Quantifier elimination achievement, commercially also has and widely uses.Current user is accessing the multimedia networks such as electric business website, Video Music
Stand so that during question and answer class website, associated recommendation can be obtained, eliminate repeat search, the uninteresting work of screening, lifting effect
Consumer's Experience is also improved while rate.For electric business website, personalized recommendation allows user to be more likely to purchase association
Product, improves sales achievement, for the content class such as video website, and personalized recommendation can then make user be interested in browse more
Many contents and then the residence time for improving user.So far, for commending system research more rest on raising proposed algorithm
The stage of efficiency or accuracy, and the essence for ignoring user is a people, everyone can produce data in multiple websites, because
This cross-domain commending system that can merge multiple website numeric field datas becomes new study hotspot.Traditional personalized recommendation system
Only recommended in particular area, when only user provides enough data in the current domain, commending system can just be given
Go out accurate recommendation.This generates the problem of above-mentioned cold start-up, Deta sparseness.With mobile Internet
Development, people can log on the mobile devices such as mobile phone, flat board and use various network services daily, such as user can use
Wechat is chatted, and checks circle of friends information flow, thumb up, comment;Also microblogging website can be logged in and check popular microblogging, thumb up, comment, hair
The new microblogging of cloth.The user registered respectively in multiple domains on network, it is probably same person to correspond in reality, this use
Family is properly termed as the intersection user between multiple domains.In practice, many business commending systems are all based on large-scale dataset
, cause, user-project matrix for collaborative filtering becomes very huge and very sparse, and this brings to commending system
Huge challenge.The typical problem that Deta sparseness is brought is exactly cold start-up problem.The recommendation of collaborative filtering is base
In the preference of user's history, new user needs just to can ensure that system is captured exactly after scoring to enough projects
Their preference, could so provide reliable recommendation.Similarly, the problem of new projects also have same.When new project is added
, it is necessary to obtain substantial amounts of user's scoring and then can just recommend suitable user after to system.
In summary, the problem of prior art is present be:Traditional personalized recommendation system is only carried out in particular area
Recommendation generates cold start-up, Deta sparseness;User-project matrix for collaborative filtering becomes very huge and very dilute
Dredge.
The content of the invention
The problem of existing for prior art, the invention provides a kind of based on the cross-domain recommendation method for intersecting user.
The present invention is achieved in that a kind of cross-domain recommendation method based on intersection user, described based on intersection user's
Cross-domain recommendation method is handled the data of source domain and aiming field, filters out intersection user, and will intersect user two
The data fusion in individual domain, constitutes a new rating matrix;Then fraction homogenization is handled, it is public using user's Similarity Measure
Formula is calculated;
Because in two fields, score-system may be different, such as source domain is 1~5 scoring, but target is led
Domain is hundred-mark system scoring, therefore carries out homogenization processing using each scoring maximum m.
User's calculating formula of similarity should be:
Wherein, the similarity simil (u, v) between user u, u ' u, v is intersected;Represent that targeted customer u scoring is averaged
Value, rviRepresent marking of the user v to project i, Iuu′Represent the common factor of two user's scorings.
Further, the rating matrix R that should be based on the cross-domain recommendation method input data for intersecting user in two domainsS
And RT, and the targeted customer u being present in simultaneously in two matrixes, specific method includes:
Input:Source domain rating matrix RS, aiming field rating matrix RT;
Targeted customer u;
Output:The project of n recommendation;
Step one, distribution intersects the rating matrix R of userST;
Step 2, takes out user u from the rating matrix of sourcei, search target rating matrix RTIn corresponding user;
Step 3, if finding correspondence user, is intersecting user's rating matrix RSTMiddle insertion user uiScore square in source
Score data in battle array and target rating matrix;
Step 4, repetitive cycling are Step 2: user in step 3, ergodic source rating matrix;
Step 5, calculates and intersects user uiWith targeted customer u similarity simil;
Step 6, user is sorted from high to low according to similarity, is taken out ranking most forward n user and is put into similar use
In the group U of family;
Step 7, calculates the average score of each user in U;
Step 8, calculates targeted customer u to aiming field RTIn non-scoring item prediction scoring;
Step 9, will predict that the project after scoring is arranged from high to low by fraction, take out the most forward n project work of ranking
For recommendation results.
Further, the computing formula scored in the step 8 is:
The scoring of user and project constitutes each element r in rating matrix R, RijRepresent that user i is commented project j
Point, if user is not to certain item rating, corresponding element is 0, as follows:
Further, it is ranked up according to the similarity between each user and targeted customer u, it is targeted customer u's to take Top-N
Neighbour;
Use formula:
Predictions of the targeted customer u to destination item i is calculated to give a mark, whereinTargeted customer u grade average is represented, U is
Targeted customer u similar users set,Represent marking of the user v to project i;Prediction scoring is ranked up, constituency Top-N
It is used as final recommendation.
Another object of the present invention is to provide the mobile phone based on the cross-domain recommendation method for intersecting user described in a kind of application.
Another object of the present invention is to provide the computer based on the cross-domain recommendation method for intersecting user described in a kind of application.
Another object of the present invention is to provide the TV based on the cross-domain recommendation method for intersecting user described in a kind of application.
Another object of the present invention is to provide the intelligence based on the cross-domain recommendation method for intersecting user described in a kind of application
Wrist-watch.
Another object of the present invention is to provide the elevator based on the cross-domain recommendation method for intersecting user described in a kind of application
Building screen.
Advantages of the present invention and good effect are:Data first to source domain and aiming field are handled, and are filtered out
Intersect user, and data fusion of the user in two domains will be intersected, constitute a new rating matrix, then fraction is uniformed
Processing, is finally calculated using improved user's calculating formula of similarity of the invention.When targeted customer is new user, tradition
Proposed algorithm can only provide a user several projects at random, recommended again according to the selection of user, it is only necessary to which user is at it
There were some data in his field, it is possible to recommended well for the new user of target domain, was improving user's " cold start-up "
Have certain help in problem, and for user that is inactive, seldom scoring, can use its active area knowledge for its
Target domain provides more accurate prediction, is also had its points of course on reply Sparse sex chromosome mosaicism.
Data of the present invention by migrated userses on source platform, recommend target platform, are put down so as to improve target
The accuracy that platform is recommended, the problems such as alleviating Sparse, the cold start-up of single domain commending system.Present invention concern is not between same area
Intersect user, i.e. same person to the different user of not same area, will not seem between same area and not have related user to connect,
And use the data that user accumulates in original domain for recommending in aiming field, used in the case of Sparse to intersect
Family provides more accurate recommend.
The present invention proposes that intersect the concept of user, i.e. same person needs to register two users on two different web sites,
This user is referred to as to intersect user;Similar performance should be had on two websites by intersecting user, if institute can be identified
Some intersection users, then also will not be complete cold start-up after the user of one website platform of user's new registration, in other websites
In similar users can still provide recommended project in targeted website;User's calculating formula of similarity of single domain commending system is equal
Data in a domain are only used.The appraisement system of same area may be very different, it is impossible to directly improve commending system original
User's calculating formula of similarity-Pearson correlation coefficient, with the addition of data of the user in source domain, more accurately;
The data that the present invention can fully rely on original field calculate user's similarity, final to produce recommendation, to a certain degree
On solve the problems, such as user's cold start-up;Improvement user's calculating formula of similarity that user intersects user is proposed, is formd new
Cross-domain proposed algorithm;Based on the cross-domain proposed algorithm for intersecting user, migration intersects data of the user in source domain to aiming field
In, strengthen the data rich degree of aiming field, so as to obtain more accurate personalized recommendation result.
Brief description of the drawings
Fig. 1 is provided in an embodiment of the present invention based on the cross-domain recommendation method flow diagram for intersecting user.
Fig. 2 is that rating matrix provided in an embodiment of the present invention is expressed as RSWith the rating matrix R of aiming fieldTDo simple concatenation
Schematic diagram.
Fig. 3 is that the cross-domain recommendation method provided in an embodiment of the present invention based on intersection user realizes flow chart.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to embodiments, to the present invention
It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to
Limit the present invention.
The application principle of the present invention is explained in detail below in conjunction with the accompanying drawings.
As shown in figure 1, provided in an embodiment of the present invention comprised the following steps based on the cross-domain recommendation method for intersecting user:
S101:Distribution intersects the rating matrix of user;
S102:User is taken out from the rating matrix of source, corresponding user in target rating matrix is searched;
S103:If finding correspondence user, intersecting user's rating matrix:Middle insertion user is in source rating matrix and mesh
Mark the score data in rating matrix;
S104:User in repetitive cycling S102 and S103, ergodic source rating matrix;
S105:Calculate the similarity for intersecting user and targeted customer;
S106:User is sorted from high to low according to similarity, ranking most forward user is taken out and is put into similar users group
In;
S107:Calculate the average score of each user;
S108:Calculate the prediction scoring of non-scoring item in targeted customer's aiming field;
S109:It will predict that the project after scoring is arranged from high to low by fraction, and take out ranking most forward project as pushing away
Recommend result.
The application principle of the present invention is further described below in conjunction with the accompanying drawings.
1st, cross-area crossover is recommended
In collaborative filtering, the scoring of user and project constitutes each element r in rating matrix R, RijRepresent to use
Scorings of the family i to project j, if user is not to certain item rating, corresponding element is 0, as follows:
According to rating matrix R, the similarity simil (u, v) between two users u, v can be calculated respectively;
It is ranked up according to the similarity between each user and targeted customer u, takes the neighbour that Top-N is targeted customer u;
Use formula:
Predictions of the targeted customer u to destination item i is calculated to give a mark, whereinTargeted customer u grade average is represented, U is
Targeted customer u similar users set,Represent marking of the user v to project i;
Finally prediction scoring is ranked up, constituency Top-N is used as final recommendation.
The rating matrix that the data of source domain are constituted is expressed as R by the present inventionSWith the rating matrix R of aiming fieldTDo simple spelling
Connect, Fig. 2 can be obtained, in fig. 2, the often row of matrix represents a user, and each column represents a project.Row table above matrix
Show the user of source domain, lower section represents aiming field user;Two pieces of area datas in lower-left and upper right of matrix are all 0, represent common
User is in another domain without any data of generation;Matrix central region is intersection user, it can be seen that intersect user at two
Data are generated in domain.When calculating two similarities for intersecting user in aiming field, aiming field data deficiencies will cause
User's similarity is inaccurate, for example in fig. 2, and two intersect user u, u ' and do not have overlapping scoring record in aiming field, the two
Similarity is 0.
If in the calculating that the data of source domain are incorporated into user's similarity, originally sparse data are enriched, use
In family similarity also more closer to reality, such as Fig. 2, two intersect users and have 4 projects scored jointly in source domain, calculate
The similarity gone out is significantly better than the result in single goal domain.The present invention is based on Pearson correlation coefficient, it is proposed that a kind of improved
User's similarity calculating method, will intersect data of the user in aiming field and source domain and introduces simultaneously, obtained more accurately
User's similarity.
2nd, the calculating of user's similarity
Vectorial U (u can be obtained by extracting some user U score data out from user-project matrix1,u2,…,un), he
Scoring vector V (v with another user V1,v2,…,vn) the distance between be two users between similarity.
Vector similarity, which is calculated, has many very ripe computing formula with calculating function, in commending system related algorithm
What is be most widely used then has cosine similarity, Pearson came (Pearson) coefficient correlation, Jie Kade (Jaccard) similarity factor
And some distance metric formula such as Euclidean distance, manhatton distance, Minkowski (Minkowski) distance etc..
Euclidean distance is most common distance metric, for representing the air line distance between 2 points, for two to
Measure U (u1,u2,…,un) and V (v1,v2,…,vn), its Euclidean distance is defined as:
Manhatton distance earliest be used for measure the distance between street because street is all square because distance according to
The difference of coordinate system is calculated, in two vector U (u1,u2,…,un) and V (v1,v2,…,vn), its manhatton distance is defined as
D=| u1-v1|+|u2-v2|+…+|un-vn|;
It is a formula that Euclidean distance and manhatton distance, which are arranged, can be obtained:
This is Minkowski Distance.
Jie Kade similarity factors are as follows to calculate two vectorial similarities, formula:
Span is J ∈ [0,1], especially, if U and V are empty set, defines J (U, V)=1.From Jie Kade phases
Jie Kade distances can be obtained like coefficient, computing formula is:
Cosine similarity is widely used in the Similarity Measure of Collaborative Filtering Recommendation System, two inner product of vectors of expression
Included angle cosine.Because cosine cos θ span is [- 1,1], therefore the scope of cosine similarity is also identical.If two
Individual vectorial similarity is that two vectors of -1 explanation are antipodal;It can illustrate two if cosine similarity is 1
Vector is identical;It is considered that two vectors are not related if the value that cosine similarity is calculated is 0.For
Other values in span, then show that two vectors have more or less similitude or diversity.For vectorial U and V,
The computing formula of cosine similarity is:
For the calculating of cosine similarity similarity in collaborative filtering, by taking the collaborative filtering based on user as an example,
Two user u and u ' are calculated, their similarity is calculated, formula should be amended as follows:
Wherein Iuu′Represent the common factor of two user's scorings.
In Collaborative Filtering Recommendation Algorithm, another conventional formula is then Pearson correlation coefficient during measurement similarity.
The full name of Pearson correlation coefficient be Pearson came product away from coefficient correlation, this coefficient is widely used for the phase between measuring vector
Pass degree, calculation is to obtain two vectorial covariances and standard deviation, is then divided by, can be expressed as with mathematical formulae:
The span of Pearson correlation coefficient is [- 1,1], when correlation coefficient ρ gets 1, shows two vector U and V
It is identical;When correlation coefficient ρ gets -1;Show that two vector U and V are completely opposite;When correlation coefficient ρ gets 0, show two
Vector is uncorrelated.The average value of vector element will be subtracted when being calculated due to Pearson correlation coefficient, therefore is applied to recommendation system
Among system, when calculating the similarity of user, the height that can't be scored because of user causes the inaccurate of Similarity Measure, because
This Collaborative Filtering Recommendation Algorithm is frequently with the similarity between Pearson correlation coefficient measuring vector.It is specifically applied to commending system
When middle, computing formula is:
Wherein i is destination item, and u is targeted customer, and formula calculates targeted customer u and user u ' similarity.
For intersecting user, if the score information being introduced into source domain, grade averageIt should be:
Because in two fields, score-system may be different, such as source domain is 1~5 scoring, but target is led
Domain is hundred-mark system scoring, therefore carries out homogenization processing using each scoring maximum m.
Final calculating formula of similarity should be:
If user belongs in new registration, formula T, part is not present, and Similarity Measure places one's entire reliance upon source domain data.
3rd, arthmetic statement
It is proposed by the present invention based on intersect user cross-domain proposed algorithm emphasis improve be and meanwhile used source domain with
The data of target domain, therefore input data should be the rating matrix R in two domainsSAnd RT, and it is present in two matrixes simultaneously
In targeted customer u, specific algorithm flow is as follows:
The rating matrix R that should be based on the cross-domain recommendation method input data for intersecting user in two domainsSAnd RT, with
And it is present in the targeted customer u in two matrixes simultaneously, specific method includes:
Input:Source domain rating matrix RS, aiming field rating matrix RT;
Targeted customer u;
Output:The project of n recommendation;
Step one, distribution intersects the rating matrix R of userST;
Step 2, takes out user u from the rating matrix of sourcei, search target rating matrix RTIn corresponding user;
Step 3, if finding correspondence user, is intersecting user's rating matrix RSTMiddle insertion user uiScore square in source
Score data in battle array and target rating matrix;
Step 4, repetitive cycling are Step 2: user in step 3, ergodic source rating matrix;
Step 5, calculates and intersects user uiWith targeted customer u similarity simil;
Step 6, user is sorted from high to low according to similarity, is taken out ranking most forward n user and is put into similar use
In the group U of family;
Step 7, calculates the average score of each user in U;
Step 8, calculates targeted customer u to aiming field RTIn non-scoring item prediction scoring;
Step 9, will predict that the project after scoring is arranged from high to low by fraction, take out the most forward n project work of ranking
For recommendation results.
It is provided in an embodiment of the present invention as shown in Figure 3 based on the cross-domain recommendation method for intersecting user.
Presently preferred embodiments of the present invention is the foregoing is only, is not intended to limit the invention, all essences in the present invention
Any modification, equivalent and improvement made within refreshing and principle etc., should be included within the scope of the present invention.
Claims (9)
1. it is a kind of based on the cross-domain recommendation method for intersecting user, it is characterised in that described based on the cross-domain recommendation side for intersecting user
Method is handled the data of source domain and aiming field, filters out intersection user, and will intersect data of the user in two domains
Fusion, constitutes a new rating matrix;Then fraction homogenization is handled, counted using user's calculating formula of similarity
Calculate;
Because in two fields, score-system may be different, such as source domain is 1~5 scoring, but target domain is
Hundred-mark system scores, therefore carries out homogenization processing using each scoring maximum m;
User's calculating formula of similarity should be:
Wherein, the similarity simil (u, v) between user u, u ' u, v is intersected;Represent targeted customer u grade average, rvi
Represent marking of the user v to project i, Iuu′Represent the common factor of two user's scorings.
2. it is as claimed in claim 1 based on the cross-domain recommendation method for intersecting user, it is characterised in that described based on intersection user
Cross-domain recommendation method input data should be rating matrix R in two domainsSAnd RT, and be present in simultaneously in two matrixes
Targeted customer u, specific method includes:
Input:Source domain rating matrix RS, aiming field rating matrix RT;
Targeted customer u;
Output:The project of n recommendation;
Step one, distribution intersects the rating matrix R of userST;
Step 2, takes out user u from the rating matrix of sourcei, search target rating matrix RTIn corresponding user;
Step 3, if finding correspondence user, is intersecting user's rating matrix RSTMiddle insertion user uiIn source rating matrix and
Score data in target rating matrix;
Step 4, repetitive cycling are Step 2: user in step 3, ergodic source rating matrix;
Step 5, calculates and intersects user uiWith targeted customer u similarity simil;
Step 6, user is sorted from high to low according to similarity, is taken out ranking most forward n user and is put into similar users group
In U;
Step 7, calculates the average score of each user in U;
Step 8, calculates targeted customer u to aiming field RTIn non-scoring item prediction scoring;
Step 9, will predict that the project after scoring is arranged from high to low by fraction, take out ranking most forward n project as pushing away
Recommend result.
3. it is as claimed in claim 2 based on the cross-domain recommendation method for intersecting user, it is characterised in that to be scored in the step 8
Computing formula be:
The scoring of user and project constitutes each element r in rating matrix R, RijScorings of the user i to project j is represented, such as
Fruit user is not to certain item rating, then corresponding element is 0, as follows:
4. it is as claimed in claim 2 based on the cross-domain recommendation method for intersecting user, it is characterised in that according to each user and target
Similarity between user u is ranked up, and takes the neighbour that Top-N is targeted customer u;
Use formula:
Predictions of the targeted customer u to destination item i is calculated to give a mark, whereinTargeted customer u grade average is represented, U is target
User u similar users set,Represent marking of the user v to project i;Prediction scoring is ranked up, constituency Top-N conducts
Final recommendation.
5. the mobile phone based on the cross-domain recommendation method for intersecting user described in a kind of application Claims 1 to 4 any one.
6. the computer based on the cross-domain recommendation method for intersecting user described in a kind of application Claims 1 to 4 any one.
7. the TV based on the cross-domain recommendation method for intersecting user described in a kind of application Claims 1 to 4 any one.
8. the intelligent watch based on the cross-domain recommendation method for intersecting user described in a kind of application Claims 1 to 4 any one.
9. the elevator building screen based on the cross-domain recommendation method for intersecting user described in a kind of application Claims 1 to 4 any one
Curtain.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710188924.2A CN106951547A (en) | 2017-03-27 | 2017-03-27 | A kind of cross-domain recommendation method based on intersection user |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710188924.2A CN106951547A (en) | 2017-03-27 | 2017-03-27 | A kind of cross-domain recommendation method based on intersection user |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106951547A true CN106951547A (en) | 2017-07-14 |
Family
ID=59473161
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710188924.2A Pending CN106951547A (en) | 2017-03-27 | 2017-03-27 | A kind of cross-domain recommendation method based on intersection user |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106951547A (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108182264A (en) * | 2018-01-09 | 2018-06-19 | 武汉大学 | A kind of ranking based on cross-cutting ranking recommended models recommends method |
CN109325146A (en) * | 2018-11-12 | 2019-02-12 | 平安科技(深圳)有限公司 | A kind of video recommendation method, device, storage medium and server |
CN109523341A (en) * | 2018-10-12 | 2019-03-26 | 广西师范大学 | The cross-domain recommended method of anonymity based on block chain technology |
CN109670161A (en) * | 2017-10-13 | 2019-04-23 | 北京京东尚科信息技术有限公司 | Commodity similarity calculating method and device, storage medium, electronic equipment |
CN109711879A (en) * | 2018-12-25 | 2019-05-03 | 广东电网有限责任公司信息中心 | Power business application intelligent interactive method and system based on micro services framework |
CN110032684A (en) * | 2019-04-22 | 2019-07-19 | 山东大学 | The cross-domain Parallel Sequence recommended method of information, medium and equipment based on shared account |
CN110110209A (en) * | 2018-01-22 | 2019-08-09 | 青岛科技大学 | A kind of intersection recommended method and system based on local weighted linear regression model (LRM) |
CN110135952A (en) * | 2019-05-16 | 2019-08-16 | 深圳市梦网百科信息技术有限公司 | A kind of Method of Commodity Recommendation and system based on category similarity |
CN111291274A (en) * | 2020-03-02 | 2020-06-16 | 苏州大学 | Article recommendation method, device, equipment and computer-readable storage medium |
CN111563205A (en) * | 2020-04-26 | 2020-08-21 | 山东师范大学 | Cross-domain information recommendation method and system based on self-attention mechanism in shared account |
CN111652669A (en) * | 2020-04-15 | 2020-09-11 | 珠海小礼鱼科技有限公司 | Intelligent terminal shopping guide system and method |
CN111737427A (en) * | 2020-05-11 | 2020-10-02 | 华南理工大学 | Mu lesson forum post recommendation method integrating forum interaction behavior and user reading preference |
CN111966914A (en) * | 2020-10-26 | 2020-11-20 | 腾讯科技(深圳)有限公司 | Content recommendation method and device based on artificial intelligence and computer equipment |
CN112035743A (en) * | 2020-08-28 | 2020-12-04 | 腾讯科技(深圳)有限公司 | Data recommendation method and device, computer equipment and storage medium |
CN112231586A (en) * | 2020-12-15 | 2021-01-15 | 平安科技(深圳)有限公司 | Course recommendation method, device, equipment and medium based on transfer learning |
CN112417288A (en) * | 2020-11-25 | 2021-02-26 | 南京大学 | Task cross-domain recommendation method for crowdsourcing software testing |
CN116501976A (en) * | 2023-06-25 | 2023-07-28 | 浙江天猫技术有限公司 | Data recommendation, model training, similar user analysis methods, apparatus and media |
-
2017
- 2017-03-27 CN CN201710188924.2A patent/CN106951547A/en active Pending
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109670161A (en) * | 2017-10-13 | 2019-04-23 | 北京京东尚科信息技术有限公司 | Commodity similarity calculating method and device, storage medium, electronic equipment |
CN109670161B (en) * | 2017-10-13 | 2023-01-31 | 北京京东尚科信息技术有限公司 | Commodity similarity calculation method and device, storage medium and electronic equipment |
CN108182264B (en) * | 2018-01-09 | 2022-04-01 | 武汉大学 | Ranking recommendation method based on cross-domain ranking recommendation model |
CN108182264A (en) * | 2018-01-09 | 2018-06-19 | 武汉大学 | A kind of ranking based on cross-cutting ranking recommended models recommends method |
CN110110209A (en) * | 2018-01-22 | 2019-08-09 | 青岛科技大学 | A kind of intersection recommended method and system based on local weighted linear regression model (LRM) |
CN109523341A (en) * | 2018-10-12 | 2019-03-26 | 广西师范大学 | The cross-domain recommended method of anonymity based on block chain technology |
CN109325146A (en) * | 2018-11-12 | 2019-02-12 | 平安科技(深圳)有限公司 | A kind of video recommendation method, device, storage medium and server |
CN109711879A (en) * | 2018-12-25 | 2019-05-03 | 广东电网有限责任公司信息中心 | Power business application intelligent interactive method and system based on micro services framework |
CN110032684A (en) * | 2019-04-22 | 2019-07-19 | 山东大学 | The cross-domain Parallel Sequence recommended method of information, medium and equipment based on shared account |
CN110135952B (en) * | 2019-05-16 | 2022-07-19 | 深圳市梦网视讯有限公司 | Commodity recommendation method and system based on class similarity |
CN110135952A (en) * | 2019-05-16 | 2019-08-16 | 深圳市梦网百科信息技术有限公司 | A kind of Method of Commodity Recommendation and system based on category similarity |
CN111291274A (en) * | 2020-03-02 | 2020-06-16 | 苏州大学 | Article recommendation method, device, equipment and computer-readable storage medium |
CN111652669A (en) * | 2020-04-15 | 2020-09-11 | 珠海小礼鱼科技有限公司 | Intelligent terminal shopping guide system and method |
CN111563205A (en) * | 2020-04-26 | 2020-08-21 | 山东师范大学 | Cross-domain information recommendation method and system based on self-attention mechanism in shared account |
CN111737427A (en) * | 2020-05-11 | 2020-10-02 | 华南理工大学 | Mu lesson forum post recommendation method integrating forum interaction behavior and user reading preference |
CN111737427B (en) * | 2020-05-11 | 2024-03-22 | 华南理工大学 | Method for recommending lesson forum posts by combining forum interaction behaviors and user reading preference |
CN112035743A (en) * | 2020-08-28 | 2020-12-04 | 腾讯科技(深圳)有限公司 | Data recommendation method and device, computer equipment and storage medium |
CN112035743B (en) * | 2020-08-28 | 2021-10-15 | 腾讯科技(深圳)有限公司 | Data recommendation method and device, computer equipment and storage medium |
CN111966914A (en) * | 2020-10-26 | 2020-11-20 | 腾讯科技(深圳)有限公司 | Content recommendation method and device based on artificial intelligence and computer equipment |
CN112417288A (en) * | 2020-11-25 | 2021-02-26 | 南京大学 | Task cross-domain recommendation method for crowdsourcing software testing |
CN112417288B (en) * | 2020-11-25 | 2024-04-12 | 南京大学 | Task cross-domain recommendation method for crowdsourcing software test |
CN112231586A (en) * | 2020-12-15 | 2021-01-15 | 平安科技(深圳)有限公司 | Course recommendation method, device, equipment and medium based on transfer learning |
CN116501976A (en) * | 2023-06-25 | 2023-07-28 | 浙江天猫技术有限公司 | Data recommendation, model training, similar user analysis methods, apparatus and media |
CN116501976B (en) * | 2023-06-25 | 2023-11-17 | 浙江天猫技术有限公司 | Data recommendation, model training, similar user analysis methods, apparatus and media |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106951547A (en) | A kind of cross-domain recommendation method based on intersection user | |
Sohail et al. | Feature extraction and analysis of online reviews for the recommendation of books using opinion mining technique | |
Bentley | Gender differences and factors affecting publication productivity among Australian university academics | |
CN107633430A (en) | A kind of Method of Commodity Recommendation based on community of colony | |
CN103473354A (en) | Insurance recommendation system framework and insurance recommendation method based on e-commerce platform | |
CN109711925A (en) | Cross-domain recommending data processing method, cross-domain recommender system with multiple auxiliary domains | |
CN102968506A (en) | Personalized collaborative filtering recommendation method based on extension characteristic vectors | |
CN106062743A (en) | Systems and methods for keyword suggestion | |
US10264082B2 (en) | Method of producing browsing attributes of users, and non-transitory computer-readable storage medium | |
Zhang et al. | CRATS: An LDA-based model for jointly mining latent communities, regions, activities, topics, and sentiments from geosocial network data | |
Zhang et al. | Proposing a new friend recommendation method, FRUTAI, to enhance social media providers' performance | |
Karbhari et al. | Recommendation system using content filtering: A case study for college campus placement | |
Kuanr et al. | Recent challenges in recommender systems: a survey | |
Pham et al. | Sentiment analysis and user similarity for social recommender system: An experimental study | |
Sun et al. | Combining online news articles and web search to predict the fluctuation of real estate market in big data context | |
Zhao et al. | Academic social network-based recommendation approach for knowledge sharing | |
Aliannejadi et al. | User model enrichment for venue recommendation | |
Liang et al. | SMS: A framework for service discovery by incorporating social media information | |
Lessa et al. | Filtering graduate courses based on LinkedIn profiles | |
Joshi et al. | Matchmaking using fuzzy analytical hierarchy process, compatibility measure and stable matching for online matrimony in India | |
Cao et al. | A Recommendation Approach Based on Product Attribute Reviews: Improved Collaborative Filtering Considering the Sentiment Polarity. | |
Rao et al. | Product recommendation system from users reviews using sentiment analysis | |
CN104462597B (en) | A kind of positive negativity of synthetic user scores and the collaborative filtering method of scoring preference heterogeneity | |
Hajikhani et al. | Brand analysis in social network services: Results from content analysis in twitter regarding the us smartphone market | |
Kaur et al. | Learner-Centric Hybrid Filtering-Based Recommender System for Massive Open Online Courses |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170714 |
|
WD01 | Invention patent application deemed withdrawn after publication |