CN108460145A - A kind of collaborative filtering recommending method based on mixing Interest Similarity - Google Patents
A kind of collaborative filtering recommending method based on mixing Interest Similarity Download PDFInfo
- Publication number
- CN108460145A CN108460145A CN201810212124.4A CN201810212124A CN108460145A CN 108460145 A CN108460145 A CN 108460145A CN 201810212124 A CN201810212124 A CN 201810212124A CN 108460145 A CN108460145 A CN 108460145A
- Authority
- CN
- China
- Prior art keywords
- user
- article
- scoring
- similarity
- interest
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
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 proposes a kind of new hybrid subscriber Interest Similarity computational methods.By establishing rating matrix of the user to project used, when finding that user's rating matrix is sky, calculates user characteristics attribute similarity searching similar users and carry out prediction scoring.When the article number that scores jointly between target user and other users is fewer, by calculating article similarity, user interest similarity is calculated indirectly.User interest similarity calculation is largely divided into three parts:It directly calculates the distance value of user's scoring, find out whether the contribution margin of one group of scoring and this group scoring are singular value in entire points-scoring system.Finally, the method for three calculating user interest similarities is realized that calculate similarity according to user property under cold start seamlessly transits to obtain end user's Interest Similarity to according to user's score information by sigmoid functions.It is scored according to the prediction of the non-scoring item of user interest similarity calculation target user, chooses the wherein highest N number of project recommendation of prediction scoring, the present invention can be effectively relieved cold start-up problem, data sparsity problem and effectively improve the accuracy that prediction is recommended.
Description
(1) technical field
The present invention relates to personalized recommendation technical fields, and in particular to a kind of collaborative filtering based on mixing Interest Similarity
The method of recommendation.
(2) background technology
Commending system is a kind of intelligence system solving information overload.Collaborative filtering be recommended technology now method it
One, its article that may like directly is predicted according to the historical behavior of user, is widely used in almost all of large-scale electricity
In sub- business web site.And there is no any scoring to record when user has just enter into system, and the article to score jointly between user
Quantity is very few to lead to data sparsity problem, and it is not accurate enough to go out user's similarity by traditional similarity calculation, passes through at this time
Hybrid similarity based on user interest can then solve the problems, such as Deta sparseness and extraction user interest.
The key technology of user interest similarity includes the calculation to user interest similitude, inaccurate similitude
It is inaccurate that calculating will cause user interest to propose, and then generate a large amount of mistake and recommend.Traditional similarity calculation mode exists
It can lead to prediction result accuracy decline when encountering Sparse Problem in system log.Current research for this problem propose from
User interest similarity is extracted in article similarity indirectly, by considering that user's score information calculates user's similitude comprehensively.But
It is the transition problem for ignoring article from sparse to intensive.Carrying out user interest Similarity measures accordingly, there may be certain inclined
Difference.Therefore it set forth herein a kind of method of the collaborative filtering based on mixing Interest Similarity, scores article according between user
Three absolute distance, contribution margin and singular value combined factors consider that user's score information calculates the interest similarity between user.
Then user interest similitude is divided into starting stage, Sparse sexual stage and terminal stage, is scored article according to user
Number three phases are mixed.This method, which effectively improves, obtains user interest accuracy rate, ensures to recommend quality.
(3) invention content
The object of the present invention is to provide a kind of recommendation methods based on mixing Interest Similarity collaborative filtering.This method passes through
User interest similarity is divided into three phases, user first has just enter into system and carries out similarity calculation according to user property, so
User's similarity, ideal user and phase are calculated according to article similarity indirectly when the number that scores jointly between user afterwards is less
User interest similarity is calculated according to the article to score jointly between user when the number that scores jointly like user is not sparse.This side
The more acurrate extraction user interest of method effectively solves cold start-up, data sparsity problem, realizes effective recommendation of commending system.
In order to achieve the above object, the present invention is realized using following scheme:
User characteristics attribute and article characteristics attribute are extracted first.The user data that commending system is newly added is sky, root
According to user characteristics attribute user's similitude is calculated to recommend, with user to article scoring quantity rise, at this time user with
The number of articles that scores jointly between user is very few, then acquires user interest similitude indirectly according to article similitude, when user it
Between score when number of articles increases to certain amount and then can directly be counted according to the article to score jointly between user jointly
Calculate user's similitude.The similar users that three kinds of user interest similitudes find target user are merged by using sigmoid functions
To predict user to the scoring for the article that scores, chooses TOP-N item lists and complete to recommend.
Following parameter is arrived involved in the present invention:
User characteristics attribute vector Attruser=(au1,au2,...,aun), ruiIndicate scorings of the user u to article i;Object
Product characteristic attribute vector is expressed as Attritem=(ai1,ai2,...,ain);ru,pIndicate scorings of the user v to article p;μpIt indicates
The average value that all users score to article P;rmedTo the intermediate value of score value in expression system;Sim (i, j) indicates article i
Similarity between j;CuvIndicate that user u and v score the set of article jointly;rupIndicate that prediction user u comments article p
Point;rkpIndicate scorings of the user k to article p;
Its specific method step is:
(1) by daily record data system, interest information of the user to each article is obtained, according to the point system of setting,
Establish rating matrix of each user to all items;
(2) average score of each user to article is calculated according to rating matrix, by user characteristics attribute and article characteristics
Attribute is indicated with vector;
(3) there is no historical data when new user enters system, obtain user property feature vector calculate it is similar between user
Spend simattr(u,v);
(4) distance value of the two in scoring is found out to the scoring of article according to two users, it is as follows calculates publicity
(5) contribution margin of one group of scoring is obtained with the gap of rating system median to article scoring according to two users,
Calculation formula is as follows:
(6) it is to judge whether one group of scoring is singular value, is compared according to this group scoring and the average value of all scorings,
Calculation formula is as follows:
(7) user interest similarity is obtained in conjunction with (6) three factors of step (4) (5), calculation formula is as follows:
sim(u,v)1 PSS=∑p∈Iproximity(ru,p,rv,p)×significance(ru,p,rv,p)×sin
gularity(ru,p,rv,p);
(8) new user enters in system, and due to no any data, take steps (3) searching similar users;
(9) when the number for the article that scores jointly between target user and other users is less, by between calculating article
Interest Similarity sim of the similarity combination step (7) when obtaining sparse between user2;
(10) as the number that scores jointly between user increases, pass through the object mainly by scoring jointly between user at this time
Product carry out user's Similarity measures;
(11) step (7) (8) (9) by user to score article number and user between score jointly article number into
Row is excessive;
(12) three kinds of similarities are merged to obtain final hybrid similarity, calculation formula is as follows:
simfinal(u, v)=α simattr(u,v)+βsim1+λsim2;
λ=1- alpha-betas;
(13) user interest similarity is obtained by step 13 and finally predicts scoring of the user to the article not scored, such as
Under:
(14) it chooses prediction and scores highest top n article to recommended user, algorithm terminates.
Recommendation method set forth in the present invention based on hybrid subscriber Interest Similarity collaborative filtering,.
The novelty of this method is:
1. when user does not just have any data to solve the problems, such as user's cold start-up according to user property similitude into system.
2. being inclined to similarity according to user's scoring similarity and user obtains user interest similarity respectively from user behavior
Similitude and user acquire user interest similitude to the tendency similitude of a certain article.
3. indirect by article characteristics attribute similarity when the article number that scores jointly between user is less according to article
It calculates user's similitude and solves data sparsity problem.
4. being divided into three phases from user property similitude to user interest Similarity measures according to user rating article
Number carries out smoothly, more acurrate to get user interest similarity.
(4) it illustrates
The schematic diagram of Fig. 1, methods described herein
Fig. 2, mixing Interest Similarity collaborative filtering recommending method flow chart
(5) specific implementation mode
It illustrates below in conjunction with the accompanying drawings and the present invention is described in more detail:
The method of the invention is characterized in that:
Whether more than confidence level user interest similarity is calculated separately according to the number for evaluating article between user jointly, when
Evaluated jointly between user article number it is sparse when, according to the similitude of the article of user's evaluation carry out similitude indirect meter
It calculates, when number is not when sparse, is then scored similarity and interest tendency degree mixing calculating user interest similarity according to user.Such as
The new user of fruit one does not score to any article when entering system, and similar use is selected by calculating user property similitude
Recommended at family.
(1) by daily record data system, interest information of the user to each article is obtained, according to the point system of setting,
Establish rating matrix of each user to all items;
(2) average score of each user to article is calculated according to rating matrix, by user characteristics attribute and article characteristics
Attribute is indicated with vector;
(3) there is no historical data when new user enters system, obtain user property feature vector calculate it is similar between user
Spend simattr(u,v);
(4) distance value of the two in scoring is found out to the scoring of article according to two users, it is as follows calculates publicity
(5) contribution margin of one group of scoring is obtained with the gap of rating system median to article scoring according to two users,
Calculation formula is as follows:
(6) it is to judge whether one group of scoring is singular value, is compared according to this group scoring and the average value of all scorings,
Calculation formula is as follows:
(7) user interest similarity is obtained in conjunction with (6) three factors of step (4) (5), calculation formula is as follows:
sim(u,v)1 PSS=∑p∈Iproximity(ru,p,rv,p)×significance(ru,p,rv,p)×sin
gularity(ru,p,rv,p);
(8) new user enters in system, and due to no any data, take steps (3) searching similar users;
(9) when the number for the article that scores jointly between target user and other users is less, by between calculating article
Interest Similarity sim of the similarity combination step (7) when obtaining sparse between user2;
(10) as the number that scores jointly between user increases, pass through the object mainly by scoring jointly between user at this time
Product carry out user's Similarity measures;
(11) step (7) (8) (9) by user to score article number and user between score jointly article number into
Row is excessive;
(12) three kinds of similarities are merged to obtain final hybrid similarity, calculation formula is as follows:
simfinal(u, v)=α simattr(u,v)+βsim1+λsim2;
λ=1- alpha-betas;
(13) user interest similarity is obtained by step 13 and finally predicts scoring of the user to the article not scored, such as
Under:
(14) it chooses prediction and scores highest top n article to recommended user, algorithm terminates.
The history scoring record of record user in systems, it is common between the article number to score and user according to user
The article number that scores selection calculates the mode of user interest similarity.When new user enters system, article is grasped without any
The record of work is needed to be calculated by user property similarity at this time, be pushed away according to the article that similar users are had higher rating
It recommends.When user has in systems there are a certain number of evaluations record, but the common scoring article between other users compared with
It is few, at this moment need the similarity according to the article evaluated between user to find out user interest phase knowledge and magnanimity indirectly, when total between user
Directly being scored according to user not when sparse with scoring article, with user interest tendency degree to calculate user interest similar for similarity
Degree.Prediction according to the user interest similarity calculation target user finally mixed to the article that scores, it is final to choose TOP-N articles
List is recommended to user.
Claims (1)
1. a kind of collaborative filtering recommending method based on mixing Interest Similarity, the invention is realized in this way:
The history scoring record of record user in systems, extracts user characteristics attribute and article characteristics attribute first.Recommend
The user data that system is newly added is sky, calculates user's similitude according to user characteristics attribute and recommends, with user couple
Article scoring quantity rises, and the number of articles that scores jointly between user and user at this time is very few, then indirect according to article similitude
Acquire user interest similitude, when the number of articles that scores jointly between user increases to certain amount then can directly according to
The article to score jointly between family carries out calculating user's similitude.User interest similitude considers user to article according to comprehensive
Scoring is calculated from three scoring distance value, contribution margin and singular value factors.Three are merged finally by sigmoid functions are used
Kind user interest similitude finds the similar users of target user to predict user to the scoring for the article that scores, and chooses
TOP-N item lists are completed to recommend.Following parameter is arrived involved in the present invention:
User characteristics attribute vector Attruser=(au1,au2,...,aun), ruiIndicate scorings of the user u to article i;Article is special
Sign attribute vector is expressed as Attritem=(ai1,ai2,...,ain);ru,pIndicate scorings of the user v to article p;μpIndicate all
The average value that user scores to article P;rmedTo the intermediate value of score value in expression system;Sim (i, j) indicates article i and j
Between similarity;CuvIndicate that user u and v score the set of article jointly;rupIndicate scorings of the prediction user u to article p;
rkpIndicate scorings of the user k to article p;Its specific method step is:
(1) by daily record data system, interest information of the user to each article is obtained, according to the point system of setting, is established
Rating matrix of each user to all items;
(2) average score of each user to article is calculated according to rating matrix, by user characteristics attribute and article characteristics attribute
It is indicated with vector;
(3) there is no historical data when new user enters system, obtain user property feature vector and calculate similarity between user
simattr(u,v);
(4) distance value of the two in scoring is found out to the scoring of article according to two users, it is as follows calculates publicity
(5) contribution margin for obtaining one group of scoring with the gap of rating system median to article scoring according to two users, calculates
Formula is as follows:
(6) it is to judge whether one group of scoring is singular value, is compared, calculated according to this group scoring and the average value of all scorings
Formula is as follows:
(7) user interest similarity is obtained in conjunction with (6) three factors of step (4) (5), calculation formula is as follows:
sim(u,v)1 PSS=∑p∈Iproximity(ru,p,rv,p)×significance(ru,p,rv,p)×singularity
(ru,p,rv,p);
(8) new user enters in system, and due to no any data, take steps (3) searching similar users;
(9) when the number for the article that scores jointly between target user and other users is less, by calculating the phase between article
Interest Similarity sim when combining step (7) to obtain sparse like degree between user2;
(10) as the number that scores jointly between user increases, at this time by mainly by the article that scores jointly between user into
Row user's Similarity measures;
(11) by user, the article number that scores jointly between the number and user of the article that scores carried out step (7) (8) (9)
Degree;
(12) three kinds of similarities are merged to obtain final hybrid similarity, calculation formula is as follows:
simfinal(u, v)=α simattr(u,v)+βsim1+λsim2;
λ=1- alpha-betas;
(13) user interest similarity is obtained by step 13 and finally predicts scoring of the user to the article not scored, it is as follows:
(14) it chooses prediction and scores highest top n article to recommended user, algorithm terminates.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810212124.4A CN108460145B (en) | 2018-03-15 | 2018-03-15 | Collaborative filtering recommendation method based on mixed interest similarity |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810212124.4A CN108460145B (en) | 2018-03-15 | 2018-03-15 | Collaborative filtering recommendation method based on mixed interest similarity |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108460145A true CN108460145A (en) | 2018-08-28 |
CN108460145B CN108460145B (en) | 2020-07-03 |
Family
ID=63216925
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810212124.4A Active CN108460145B (en) | 2018-03-15 | 2018-03-15 | Collaborative filtering recommendation method based on mixed interest similarity |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108460145B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109697628A (en) * | 2018-11-12 | 2019-04-30 | 平安科技(深圳)有限公司 | Product data method for pushing and device, storage medium, computer equipment |
CN110083764A (en) * | 2019-04-11 | 2019-08-02 | 东华大学 | A kind of collaborative filtering cold start-up way to solve the problem |
CN111274493A (en) * | 2020-01-17 | 2020-06-12 | 电子科技大学 | Grading prediction method based on multi-source user comments |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112364254A (en) * | 2020-10-09 | 2021-02-12 | 天津大学 | Collaborative filtering recommendation system and method for improving user similarity |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100153292A1 (en) * | 2008-12-11 | 2010-06-17 | Microsoft Corporation | Making Friend and Location Recommendations Based on Location Similarities |
CN102609533A (en) * | 2012-02-15 | 2012-07-25 | 中国科学技术大学 | Kernel method-based collaborative filtering recommendation system and method |
CN103093376A (en) * | 2013-01-16 | 2013-05-08 | 北京邮电大学 | Clustering collaborative filtering recommendation system based on singular value decomposition algorithm |
CN104063481A (en) * | 2014-07-02 | 2014-09-24 | 山东大学 | Film individuation recommendation method based on user real-time interest vectors |
CN107391670A (en) * | 2017-07-21 | 2017-11-24 | 云南电网有限责任公司教育培训评价中心 | A kind of mixing recommendation method for merging collaborative filtering and user property filtering |
-
2018
- 2018-03-15 CN CN201810212124.4A patent/CN108460145B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100153292A1 (en) * | 2008-12-11 | 2010-06-17 | Microsoft Corporation | Making Friend and Location Recommendations Based on Location Similarities |
CN102609533A (en) * | 2012-02-15 | 2012-07-25 | 中国科学技术大学 | Kernel method-based collaborative filtering recommendation system and method |
CN103093376A (en) * | 2013-01-16 | 2013-05-08 | 北京邮电大学 | Clustering collaborative filtering recommendation system based on singular value decomposition algorithm |
CN104063481A (en) * | 2014-07-02 | 2014-09-24 | 山东大学 | Film individuation recommendation method based on user real-time interest vectors |
CN107391670A (en) * | 2017-07-21 | 2017-11-24 | 云南电网有限责任公司教育培训评价中心 | A kind of mixing recommendation method for merging collaborative filtering and user property filtering |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109697628A (en) * | 2018-11-12 | 2019-04-30 | 平安科技(深圳)有限公司 | Product data method for pushing and device, storage medium, computer equipment |
CN109697628B (en) * | 2018-11-12 | 2024-05-10 | 平安科技(深圳)有限公司 | Product data pushing method and device, storage medium and computer equipment |
CN110083764A (en) * | 2019-04-11 | 2019-08-02 | 东华大学 | A kind of collaborative filtering cold start-up way to solve the problem |
CN111274493A (en) * | 2020-01-17 | 2020-06-12 | 电子科技大学 | Grading prediction method based on multi-source user comments |
Also Published As
Publication number | Publication date |
---|---|
CN108460145B (en) | 2020-07-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108460145A (en) | A kind of collaborative filtering recommending method based on mixing Interest Similarity | |
CN102929928B (en) | Multidimensional-similarity-based personalized news recommendation method | |
CN103412938B (en) | A kind of commodity price-comparing method extracted based on picture interactive multiobjective | |
CN104199896B (en) | The video similarity of feature based classification is determined and video recommendation method | |
CN103902538B (en) | Information recommending apparatus and method based on decision tree | |
CN104935963B (en) | A kind of video recommendation method based on timing driving | |
CN107833117B (en) | Bayesian personalized sorting recommendation method considering tag information | |
CN104966125B (en) | A kind of article scoring of social networks and recommend method | |
CN102508870B (en) | Individualized recommending method in combination of rating data and label data | |
CN102411754A (en) | Personalized recommendation method based on commodity property entropy | |
CN108108380A (en) | Search ordering method, searching order device, searching method and searcher | |
CN110427567A (en) | A kind of collaborative filtering recommending method based on user preference Similarity-Weighted | |
CN104317900A (en) | Multiattribute collaborative filtering recommendation method oriented to social network | |
CN103559622A (en) | Characteristic-based collaborative filtering recommendation method | |
CN103412937A (en) | Searching and shopping method based on handheld terminal | |
CN104298749A (en) | Commodity retrieval method based on image visual and textual semantic integration | |
CN102841929A (en) | Recommending method integrating user and project rating and characteristic factors | |
CN103870516A (en) | Image retrieving method, real-time drawing prompting method and device thereof | |
CN104809243A (en) | Mixed recommendation method based on excavation of user behavior compositing factor | |
CN106326483A (en) | Collaborative recommendation method with user context information aggregation | |
CN102135999A (en) | User credibility and item nearest neighbor combination Internet recommendation method | |
CN108874916A (en) | A kind of stacked combination collaborative filtering recommending method | |
CN108876536A (en) | Collaborative filtering recommending method based on arest neighbors information | |
CN103021404A (en) | Advertisement identification method based on audio | |
CN104021230B (en) | Collaborative filtering method based on community discovery |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |