CN103514239B - A kind of integrated user behavior and the recommendation method and system of item contents - Google Patents
A kind of integrated user behavior and the recommendation method and system of item contents Download PDFInfo
- Publication number
- CN103514239B CN103514239B CN201210486275.1A CN201210486275A CN103514239B CN 103514239 B CN103514239 B CN 103514239B CN 201210486275 A CN201210486275 A CN 201210486275A CN 103514239 B CN103514239 B CN 103514239B
- Authority
- CN
- China
- Prior art keywords
- user
- article
- content
- data
- vector
- 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.)
- Active
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
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
- G06F16/337—Profile generation, learning or modification
-
- 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)
- Computational Linguistics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses the recommendation method and system of a kind of integrated user behavior and item contents.Described method includes: by associating behavioral data and the content-data of all items of each user, obtain each user user interest list to all items content；Calculate the similarity of described user interest list and the content-data of described all items, draw each user recommendation article weight to the content-data of article, and by recommending article weight to be ranked up drawing the recommendation article of each user.The present invention effectively solves the subject matter that current commending system is faced, the cold start-up of such as commending system, and it is the most excessive to normal operating condition etc. smoothly from cold start-up, substantially increase rate of accurateness, coverage rate, novelty degree etc., can preferably improve precision and the personalization of commending system, attract more user to use.
Description
Technical field
The present invention relates to intelligent recommendation system, a kind of integrated user behavior and the recommendation side of item contents
Method and system.
Background technology
Along with information technology and the development of the Internet, how from substantial amounts of information, to find the information that user is interested, as
What allows the information of oneself be welcome by users, is all an extremely difficult thing for user and provider.
The task of commending system contacts user and information exactly, on the one hand helps to the user discover that oneself valuable information, and another
Aspect allows information can be presented in face of its interesting user, thus realizes the double of information consumer and informant
Win.It mainly by analyzing the behavior of user, is modeled, thus predicts the interest of user and push away to user by commending system
Recommend.
According to the different modes processing data, it is recommended that the method that system uses can be divided mainly into collaborative filtering and content mistake
Filter, if commending system is just with the behavioral data of user, does to user according to the historical interest of user and recommends, then this
The method of kind is referred to as collaborative filtering (collaborative filtering).If commending system make use of the interior of article
Hold data (content filtering), calculate the similarity between the interest of user and article description, push away to user
Recommend, referred to as information filtering.Collaborative filtering is the foremost algorithm in commending system field, this algorithm mainly going through by research user
History behavior to the interest modeling of user and makes recommendation to user.It mainly comprises collaborative filtering based on user, based on thing
The collaborative filtering of product, and model based on theme.Information filtering is to recommend and he to user on the basis of based on item contents
Other article that before, the article liked are the most similar.Content mainly includes some attributes of article.This method can
Clearly to understand the point of interest of user, thus preferably explain the reason of recommendation.
Collaborative filtering depends on historical behavior data, the most not can solve the cold start-up problem of system, also
Recommend that is these algorithms do to new user, because there is no relevant behavioral data.Simultaneously also cannot be by new article
Recommend may be interested in it user.Information filtering algorithm comes to user mainly by the similarity of content between article
Recommending, it only depends on the data of item contents itself, so can solve cold start-up problem.But due to information filtering
Algorithm ignores user behavior, thus also ignores the rule included in the popularity of article and user behavior, so it
Ratio of precision collaborative filtering low.How to merge the advantage of both algorithms to design a kind of new proposed algorithm for recommending system
Cold start-up and the precision of raising commending system that system faces are of great significance.
Therefore, prior art has yet to be improved and developed.
Summary of the invention
The technical problem to be solved in the present invention is, for the drawbacks described above of prior art, it is provided that a kind of integrated user's row
For the recommendation method and system with item contents, solve the cold start-up and from cold start-up the most smoothly of existing commending system
Excessively to the problem of normal operating condition.
It is as follows that the present invention solves the technical scheme that technical problem used:
A kind of integrated user behavior and the recommendation method of item contents, wherein, comprise the following steps:
A, by associating the behavioral data of each user and the content-data of all items, obtain each user to property
The user interest list of product content；
B, calculate the similarity of the content-data of described user interest list and described all items, draw each user couple
The recommendation article weight of the content-data of article, and by recommending article weight to be ranked up drawing the recommendation thing of each user
Product.
Described integrated user behavior and the recommendation method of item contents, wherein, described step A specifically includes:
In A1, the behavior vector that respectively content-data of the behavioral data of user and article is used user and article
Hold vector representation, and the content vector of the behavior vector sum article by the method association user of matrix decomposition, obtain user's
Interest vector.
Described integrated user behavior and the recommendation method of item contents, wherein, described step B the most also includes:
The content-data of newly-increased article is updated by the content-data of all items, and by calculating described newly-increased article
Content-data and the similarity of the behavioral data of all users, show that the recommendation user of the behavioral data of user is weighed by newly-increased article
Weight, and by recommending article weight to be ranked up drawing the recommendation user of newly-increased article.
Described integrated user behavior and the recommendation method of item contents, wherein, calculate the emerging of user by below equation
Inclination amount:
Formula (1)
Wherein,Represent the behavior vector u of unique user,Represent the interior of Individual Items
Hold vector c,Represent the content matrix C of all items,Represent
Interest vector f, n and the m of unique user are natural number,；
When the content matrix of all items is fixed, derived the interest vector f of unique user by above-mentioned formula (1).
Described integrated user behavior and the recommendation method of item contents, wherein, when Adding User and described newly-increased use
When family produces behavioral data, the behavior vector Added User by the behavioral data Added User, and by above-mentioned formula
(1) draw the interest vector Added User, and derive, by below equation (2), the recommendation article weight vectors Added User:
Formula (2)
Wherein,Represent and recommend article weight vectors, obtained by above-mentioned recommendation article weight vectors and recommend
Article recommend user.
Described integrated user behavior and the recommendation method of item contents, wherein, when there being newly-increased article, according to described institute
The content matrix C having article updates the content vector of newly-increased article, and is derived the recommendation user of newly-increased article by formula (3)
Weight vectors:
Formula (3)
Wherein,Represent described user behavior matrix U,Represent and recommend user
Weight vectors, obtains the recommendation user of newly-increased article by above-mentioned recommendation user's weight vectors.
Described integrated user behavior and the recommendation method of item contents, wherein, described step A also includes: use sparse
The data structure of storage calculates the interest vector of unique user.
A kind of integrated user behavior and the commending system of item contents, wherein, described system includes:
Relating module, for behavioral data and the content-data of all items by associating each user, obtains each
User's user interest list to all items content；
Recommend article module, similar to the content-data of described all items for calculating described user interest list
Degree, draws each user recommendation article weight to the content-data of article, and by recommending article weight to be ranked up
Go out the recommendation article of each user.
Described integrated user behavior and the commending system of item contents, wherein, described system includes:
Recommend line module, for being updated the content-data of newly-increased article by the content-data of all items, and pass through
Calculate the similarity of the content-data of described newly-increased article and the behavioral data of all users, draw the newly-increased article row to user
For recommendation user's weight of data, and by recommending article weight to be ranked up drawing the recommendation user of newly-increased article.
Integrated user behavior provided by the present invention and the recommendation method and system of item contents, effectively solve current
The cold start-up of the subject matter that commending system is faced, such as commending system, and the most excessive to just smoothly from cold start-up
Often running status etc., substantially increases rate of accurateness, coverage rate, novelty degree etc., can preferably improve commending system
Precision and personalization, attract more user to use.
Accompanying drawing explanation
Fig. 1 is the flow chart of the recommendation method of the integrated user behavior that provides of the present invention and item contents.
Fig. 2 is the structural representation of the commending system of the integrated user behavior that provides of the present invention and item contents.
Detailed description of the invention
For making the purpose of the present invention, technical scheme and advantage clearer, clear and definite, develop simultaneously embodiment pair referring to the drawings
The present invention further describes.Should be appreciated that specific embodiment described herein, and need not only in order to explain the present invention
In limiting the present invention.
See the flow chart that Fig. 1, Fig. 1 are the recommendation methods of the integrated user behavior that provides of the present invention and item contents, bag
Include following steps:
Step S100, by associating the behavioral data of each user and the content-data of all items, obtain each user
User interest list to all items content；
Step S200, calculate the similarity of the content-data of described user interest list and described all items, draw every
The individual user recommendation article weight to the content-data of article, and by recommending article weight to be ranked up drawing each user
Recommendation article.
Below in conjunction with specific embodiment, above-mentioned steps is described in detail.
The present invention mainly carrys out the interest modeling to user by the content-data of integrated article and the behavioral data of user,
Thus user is recommended by the advantage being effectively combined information filtering and collaborative filtering.Specifically, the present invention is broadly divided into
Off-line and online two parts, off-line part primarily to obtain the interest list of user, online part be then by
The user interest list that off-line part obtains, be given when Adding User its recommend accordingly article and when newly-increased article to
Go out it and recommend user accordingly.
In off-line part, in order to obtain the interest list of user, first the interest of user is modeled, defines a theme
Model, in order to obtain the interest list of user.By content-data and the behavioral data of user of associated article, topic model profit
User's interest list to item contents is obtained with the data mining algorithm of matrix decomposition.User is obtained to thing in off-line part
The main flow of the interest list of product content is as follows: the most respectively the behavioral data of user and the content-data of article is used and is used
The behavior vector at family and the content vector representation of article, the behavior vector u of unique user is expressed as,
The behavioural matrix U=of all users, the content vector of Individual Items
C is expressed as, the content matrix of all items, single
The interest vector f of individual user usesRepresenting, such topic model is:
The behavioral data of existing subscriber and the content-data of article are obtained the interest list of user by topic model；And
If having new user behavior or the input of new item contents, then update all of user behavior data or all of article
The vector space model of content-data, so to obtain the power of its feature in the vector space model of the overall situation more accurately
Weight, then goes to obtain the user interest list after updating by topic model.
In online part, set up the recommended models of commending system, it is recommended that model is mainly weighed by the interest list of user
Amount user's fancy grade to each article, and provide recommendation results online.Here, it is recommended that model is divided into two types, one
Individual is the recommendation object model for user, and one is the recommendation user model for article.Recommended models is being defined
Time, carry out different process according to the difference of recommended models type.When user is recommended article, if existing user,
Can directly inquire the interest list of this user, be then given by recommendation object model and recommend article online；If
Add User and create corresponding behavioral data, in order to obtain this behavioral data Added User exactly in the overall situation
Weight, the behavioral data first passing through all users goes to update this behavioral data Added User, and then passes through topic model
Go obtain its user interest list and provide recommendation article.When newly-increased article recommend user, if there being the content of newly-increased article
Data, first pass through all the elements data and remove to update the content-data of these newly-increased article, then by the interest of all users
List and recommendation user model are gone to obtain and are recommended user.
Specifically, it is recommended that object model is as follows:
Recommend user model as follows:
Concrete recommendation process is: when Add User and described in Add User generation behavioral data time, by newly-increased use
The behavior vector that the behavioral data at family Adds User, and by drawing the interest vector Added User, and pushed away by following
Recommend object model and derive the recommendation article weight vectors Added User, obtained by above-mentioned recommendation article weight vectors and recommend thing
Product recommend user.When there being newly-increased article, according to the content matrix C of described all items update the content of newly-increased article to
Amount, and by recommending user model to derive recommendation user's weight vectors of newly-increased article, then by above-mentioned recommendation user's weight
Vector obtains the recommendation user of newly-increased article.
By above-mentioned two recommended models, can draw respectively recommend article weight vectors and recommend the weight of user to
Amount, according to recommending the weight vectors of article and recommending the weight vectors of user can provide recommendation article the most accurately and recommend to use
Family, recommendation article here and recommendation user are not limited to recommend one, can recommend multiple.Analyze user's by interest list
Hobby, draws preferable rationale for the recommendation.
The present invention uses matrix decomposition method to obtain the interest list of user, and concrete calculating process is as follows: the content of article
The behavioral data vector space model of data and user represents.The content weight vector of each article is represented by, the behavior vector of each user is represented by, TF-IDF(Term can be passed through
Frequency Inverse Document Frequency, the anti-document frequency of word frequency) method calculates, simultaneously by these to
Amount regularization is to avoid the error calculated；
TF(word frequency) calculate the number of times that occurs in the vector of existing object of feature, DF(document frequency) calculating whole to
During in quantity space, feature occurs in how many different objects.When having new user behavior data or new item contents data
Add, will update according to this computation model.
The interests matrix of definition user is F, in order to obtain F, and objective function J:
Wherein,It it is Frobenius norm.For asking,Can be rewritten as:
By using Lagrangian to solve this conditions object function J.BecauseMatrix has been fixed, and the most only needs
Solve.The Lagrange multiplier assuming its correspondence is, then LagrangianL is defined as:
RightLocal derviation is asked to obtain
Therefore,
.
From above-mentioned formula it can be seen that object function is monotone decreasing convergence.When initializing,Can be some with
Machine number, but in order to obtain unique user interest list, we initializeFor being all the matrix of 1, in other words obtaining
It is considered that user is consistent to all of interest before real user interest vector.Along with the constantly iteration of algorithm,Can little by little restrain.The end condition of algorithm can arrange certain iterations, or when the error precision of object function
Within the scope of certain.
The number simultaneously taking account of user and article is the hugest, and user behavior data and item contents data matrix are non-
The most sparse, reduce the consumption of internal memory in order to improve the speed of calculating simultaneously, use sparse storage mode, by sparse storage
Data structure calculates the interest vector of unique user, will value two arrays cited with it of non-zero store, the most permissible
Effectively carry out the computing of sparse matrix, solve big matrix operations, obtain recommendation list rapidly.
Above-mentioned to unique userWhen recommending article, need to calculate the interest vector of userContent with all items
MatrixBetween recommendation article weight vectors, due toWithThe most immobilization, therefore it
Product i.e. cosine similarity.By to recommending article weight vectors to be ranked up obtaining maximally related article.With
Reason, when newly-increased article are recommended user, calculates the content vector of newly-increased articleInterests matrix with all usersIt
Between recommendation user's weight vectors.Then its user being ranked up obtaining the highest several weight is pushed away
Recommend.
Further, the behavioral data of the user in the present invention can have a multiple expression-form, for example, dominant feedback or
Person's stealth is fed back, display feedback can be user to the scoring of article stealthy feedback can carry out table from the record that browses of user
Show.And the content-data of article can be with the label of article, text description of article etc..No matter which kind of representation, this
The model of bright proposition takes full advantage of various different data representation form to be recommended, and can effectively processing system from
Cold start-up is to properly functioning whole process, thus has wide applicability.
Based on above-mentioned recommendation method, present invention also offers a kind of integrated user behavior and the commending system of item contents,
As in figure 2 it is shown, described system includes:
Relating module 10, for behavioral data and the content-data of all items by associating each user, obtains every
The individual user user interest list to all items content；
Recommend article module 20, similar to the content-data of described all items for calculating described user interest list
Degree, draws each user recommendation article weight to the content-data of article, and by recommending article weight to be ranked up
Go out the recommendation article of each user.
Recommend line module 30, for being updated the content-data of newly-increased article by the content-data of all items, and lead to
Cross the similarity of the content-data calculating described newly-increased article and the behavioral data of all users, show that newly-increased article are to user's
Recommendation user's weight of behavioral data, and by recommending article weight to be ranked up drawing the recommendation user of newly-increased article.
Preferably, the commending system that the present invention provides can be widely applied to various intelligent recommendation system, such as film pushes away
Recommend system, music commending system, reading commending system etc..
In sum, the integrated user behavior of present invention offer and the recommendation method and system of item contents, pass through data
Mining model is effectively combined this behavior recommendation and the advantage of commending contents, the master that the current commending system of effective solution is faced
Want problem, the cold start-up of such as commending system, and the most excessive to normal operating condition etc. smoothly from cold start-up.This
Bright integrated user behavior and the recommendation method and system of item contents can overcome not to be had content or does not has behavior both
Special circumstances, and substantially increase rate of accurateness, coverage rate and novelty degree etc., can preferably improve the precision of commending system
And personalization, attract more user to use.
It should be appreciated that the application of the present invention is not limited to above-mentioned citing, for those of ordinary skills, can
To be improved according to the above description or to convert, all these modifications and variations all should belong to the guarantor of claims of the present invention
Protect scope.
Claims (8)
1. an integrated user behavior and the recommendation method of item contents, it is characterised in that comprise the following steps:
A, by associating the behavioral data of each user and the content-data of all items, obtain each user in all items
The user interest list held；
B, calculate the similarity of the content-data of described user interest list and described all items, show that each user is to article
The recommendation article weight of content-data, and by recommending article weight to be ranked up drawing the recommendation article of each user；
Described step A specifically includes:
A1, respectively the content-data of the behavioral data of user and article is used the behavior vector of user and the content of article to
Amount represents, and the content vector of the behavior vector sum article by the method association user of matrix decomposition, obtains the interest of user
Vector；
Described step A specifically include the behavior vector that respectively content-data of the behavioral data of user and article is used user with
And the content vector representation of article, generate the behavioural matrix of all users according to the behavior vector of unique user, according to single thing
The content vector of product generates the content matrix of all items, then the topic model of definition unique user behavior vector is all items
The inner product of the interest vector of content matrix and unique user；
The behavioral data of existing subscriber and the content-data of article are obtained the interest list of user by topic model；And if
There are new user behavior or the input of new item contents, then update all of user behavior data or all of item contents
The vector space model of data, goes to obtain the user interest list after updating by topic model.
Integrated user behavior the most according to claim 1 and the recommendation method of item contents, it is characterised in that described step
B the most also includes:
The content-data of newly-increased article is updated by the content-data of all items, and by calculating the content of described newly-increased article
Data and the similarity of the behavioral data of all users, draw newly-increased article recommendation user's weight to the behavioral data of user,
And by recommending article weight to be ranked up drawing the recommendation user of newly-increased article.
Integrated user behavior the most according to claim 1 and the recommendation method of item contents, it is characterised in that by following
The interest vector of formula calculating user:
Formula (1)
Wherein,Represent the behavior vector u of unique user,Represent the content vector of Individual Items
C,Represent the content matrix C of all items,Represent single use
Interest vector f, n and the m at family are natural number,；
When the content matrix of all items is fixed, derived the interest vector f of unique user by above-mentioned formula (1).
Integrated user behavior the most according to claim 3 and the recommendation method of item contents, it is characterised in that newly-increased when having
User and described in Add User generation behavioral data time, the behavior Added User by the behavioral data Added User to
Amount, and draw the interest vector Added User by above-mentioned formula (1), and derived by below equation (2) and Add User
Recommendation article weight vectors:
Formula (2)
Wherein,Represent and recommend article weight vectors, obtained by above-mentioned recommendation article weight vectors and recommend article
Recommend user.
Integrated user behavior the most according to claim 3 and the recommendation method of item contents, it is characterised in that newly-increased when having
During article, update the content vector of newly-increased article according to the content matrix C of described all items, and derived by formula (3)
Recommendation user's weight vectors of newly-increased article:
Formula (3)
Wherein,Represent described user behavior matrix U,Represent and recommend user's weight
Vector, obtains the recommendation user of newly-increased article by above-mentioned recommendation user's weight vectors.
Integrated user behavior the most according to claim 1 and the recommendation method of item contents, it is characterised in that described step
A also includes: use the data structure of sparse storage to calculate the interest vector of unique user.
7. an integrated user behavior and the commending system of item contents, it is characterised in that described system includes:
Relating module, for behavioral data and the content-data of all items by associating each user, obtains each user
User interest list to all items content；
Recommend article module, for calculating the similarity of described user interest list and the content-data of described all items,
Go out each user recommendation article weight to the content-data of article, and by each to recommending article weight to be ranked up to draw
The recommendation article of user；
Described relating module specifically for respectively the content-data of the behavioral data of user and article is used the behavior of user to
Amount and the content vector representation of article, and by the content of the behavior vector sum article of the method for matrix decomposition association user to
Amount, obtains the interest vector of user；
Respectively the behavioral data of user and the content-data of article are used behavior vector and the content vector of article of user
Represent, generate the behavioural matrix of all users according to the behavior vector of unique user, generate according to the content vector of Individual Items
The content matrix of all items, then the topic model of definition unique user behavior vector is all items content matrix and single use
The inner product of the interest vector at family；
The behavioral data of existing subscriber and the content-data of article are obtained the interest list of user by topic model；And if
There are new user behavior or the input of new item contents, then update all of user behavior data or all of item contents
The vector space model of data, goes to obtain the user interest list after updating by topic model.
Integrated user behavior the most according to claim 7 and the commending system of item contents, it is characterised in that described system
Including:
Recommend line module, for being updated the content-data of newly-increased article by the content-data of all items, and by calculating
The content-data of described newly-increased article and the similarity of the behavioral data of all users, draw the newly-increased article behavior number to user
According to recommendation user's weight, and by recommend article weight be ranked up drawing the recommendation user of newly-increased article.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210486275.1A CN103514239B (en) | 2012-11-26 | 2012-11-26 | A kind of integrated user behavior and the recommendation method and system of item contents |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210486275.1A CN103514239B (en) | 2012-11-26 | 2012-11-26 | A kind of integrated user behavior and the recommendation method and system of item contents |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103514239A CN103514239A (en) | 2014-01-15 |
| CN103514239B true CN103514239B (en) | 2016-12-21 |
Family
ID=49896970
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210486275.1A Active CN103514239B (en) | 2012-11-26 | 2012-11-26 | A kind of integrated user behavior and the recommendation method and system of item contents |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103514239B (en) |
Families Citing this family (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104063589B (en) * | 2014-06-16 | 2018-01-16 | 百度移信网络技术（北京）有限公司 | A kind of recommendation method and system |
| CN105574025B (en) * | 2014-10-15 | 2018-10-16 | 阿里巴巴集团控股有限公司 | For calculating sequence point and establishing the method, apparatus and commercial product recommending system of model |
| CN104766219B (en) * | 2015-03-19 | 2018-05-22 | 中国船舶重工集团公司第七0九研究所 | Based on the user's recommendation list generation method and system in units of list |
| CN106326277B (en) * | 2015-06-30 | 2019-07-16 | 上海证大喜马拉雅网络科技有限公司 | A kind of audio personalization recommended method and system based on user behavior |
| CN106959966A (en) * | 2016-01-12 | 2017-07-18 | 腾讯科技（深圳）有限公司 | A kind of information recommendation method and system |
| CN105787055B (en) * | 2016-02-26 | 2020-04-21 | 合一网络技术(北京)有限公司 | Information recommendation method and device |
| CN107665208B (en) * | 2016-07-28 | 2019-12-13 | 北京国双科技有限公司 | User preference measurement method and device |
| CN106599047B (en) * | 2016-11-10 | 2020-08-18 | 咪咕动漫有限公司 | Information pushing method and device |
| CN106874374A (en) * | 2016-12-31 | 2017-06-20 | 杭州益读网络科技有限公司 | A kind of recommendation method for pushing based on user's history behavior interaction analysis |
| CN106651542B (en) * | 2016-12-31 | 2021-06-25 | 珠海市魅族科技有限公司 | Article recommendation method and device |
| CN107239993B (en) * | 2017-05-24 | 2020-11-24 | 海南大学 | Matrix decomposition recommendation method and system based on expansion label |
| CN108038217B (en) * | 2017-12-22 | 2021-05-11 | 北京星选科技有限公司 | Information recommendation method and device |
| CN109389497A (en) * | 2018-09-05 | 2019-02-26 | 平安科技（深圳）有限公司 | Intelligent recommendation method, apparatus, computer equipment and the storage medium of product sale |
| CN109753275B (en) * | 2018-12-10 | 2022-04-12 | 东软集团股份有限公司 | Recommendation method and device for application programming interface, storage medium and electronic equipment |
| CN109829116B (en) * | 2019-02-14 | 2021-07-30 | 北京达佳互联信息技术有限公司 | Content recommendation method and device, server and computer readable storage medium |
| CN110008408B (en) * | 2019-04-12 | 2021-04-06 | 山东大学 | Session recommendation method, system, device and medium |
| CN110223107B (en) * | 2019-05-23 | 2021-12-07 | 中国银行股份有限公司 | Reference advertisement determination method, device and equipment based on similar objects |
| CN112312216B (en) * | 2020-10-16 | 2022-08-16 | 山东海看新媒体研究院有限公司 | Traceable television recommendation method and system based on modular factorial theory |
| CN112559856B (en) * | 2020-12-04 | 2023-06-02 | 中国联合网络通信集团有限公司 | Application ordering method and device |
| CN113744021A (en) * | 2021-02-08 | 2021-12-03 | 北京沃东天骏信息技术有限公司 | Recommendation method, recommendation device, computer storage medium and recommendation system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101673286A (en) * | 2008-09-08 | 2010-03-17 | 索尼株式会社 | Apparatus, method and computer program for content recommendation and recording medium |
| CN101923545A (en) * | 2009-06-15 | 2010-12-22 | 北京百分通联传媒技术有限公司 | Method for recommending personalized information |
| CN102426686A (en) * | 2011-09-29 | 2012-04-25 | 南京大学 | Internet information product recommending method based on matrix decomposition |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012093046A2 (en) * | 2011-01-05 | 2012-07-12 | Thomson Licensing | Hybrid content recommendation system using matrices breakdowns |
-
2012
- 2012-11-26 CN CN201210486275.1A patent/CN103514239B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101673286A (en) * | 2008-09-08 | 2010-03-17 | 索尼株式会社 | Apparatus, method and computer program for content recommendation and recording medium |
| CN101923545A (en) * | 2009-06-15 | 2010-12-22 | 北京百分通联传媒技术有限公司 | Method for recommending personalized information |
| CN102426686A (en) * | 2011-09-29 | 2012-04-25 | 南京大学 | Internet information product recommending method based on matrix decomposition |
Non-Patent Citations (2)
| Title |
|---|
| Joining Collaborative and Content-based Filtering;Patrick Baudisch;《the ACM CHI Workshopon Interacting with Recommender Systems》;19990531;第1-4页 * |
| 协同过滤***项目冷启动的混合推荐算法;郭艳红 等;《计算机工程》;20081231;第34卷(第23期);第11-13页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103514239A (en) | 2014-01-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103514239B (en) | A kind of integrated user behavior and the recommendation method and system of item contents | |
| Li et al. | Deep collaborative filtering via marginalized denoising auto-encoder | |
| Khan et al. | CNN with depthwise separable convolutions and combined kernels for rating prediction | |
| Li et al. | Learning from history and present: Next-item recommendation via discriminatively exploiting user behaviors | |
| Ma et al. | Learning to recommend with trust and distrust relationships | |
| US8352473B2 (en) | Product synthesis from multiple sources | |
| Hu et al. | Movie collaborative filtering with multiplex implicit feedbacks | |
| CN106600302A (en) | Hadoop-based commodity recommendation system | |
| CN109522474A (en) | Recommended method based on interaction sequence data mining depth user's similitude | |
| CN103399858A (en) | Socialization collaborative filtering recommendation method based on trust | |
| CN104063481A (en) | Film individuation recommendation method based on user real-time interest vectors | |
| Ahmed et al. | Latent factor models with additive and hierarchically-smoothed user preferences | |
| WO2014100321A2 (en) | Feature embedding in matrix factorization | |
| Zheng et al. | An unexpectedness-augmented utility model for making serendipitous recommendation | |
| CN109190030A (en) | Implicit feedback recommendation method fusing node2vec and deep neural network | |
| Phuong et al. | Graph-based context-aware collaborative filtering | |
| Kavinkumar et al. | A hybrid approach for recommendation system with added feedback component | |
| CN103365842B (en) | A kind of page browsing recommends method and device | |
| CN103150667A (en) | Individualization recommending method based on mechanical structure | |
| Misztal-Radecka et al. | Meta-User2Vec model for addressing the user and item cold-start problem in recommender systems | |
| CN105045865A (en) | Kernel-based collaborative theme regression tag recommendation method | |
| Wang et al. | BPF++: a unified factorization model for predicting retweet behaviors | |
| US20140214751A1 (en) | Mixed collaborative filtering-content analysis model | |
| CN108572981B (en) | Information recommendation method and device | |
| Yu et al. | A Cross‐Domain Collaborative Filtering Algorithm Based on Feature Construction and Locally Weighted Linear Regression |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |