CN101826114B - Multi Markov chain-based content recommendation method - Google Patents
Multi Markov chain-based content recommendation method Download PDFInfo
- Publication number
- CN101826114B CN101826114B CN2010101828444A CN201010182844A CN101826114B CN 101826114 B CN101826114 B CN 101826114B CN 2010101828444 A CN2010101828444 A CN 2010101828444A CN 201010182844 A CN201010182844 A CN 201010182844A CN 101826114 B CN101826114 B CN 101826114B
- Authority
- CN
- China
- Prior art keywords
- user
- markov model
- model
- markov
- information
- 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.)
- Expired - Fee Related
Links
Images
Abstract
The invention discloses a multi Markov chain-based content recommendation method. The method comprises the following steps of: establishing Markov models by using information of click stream of a user and establishing a user relationship matrix by using the background information of the user; combining the similar Markov models; and filling sparse items in the zero line of the combined Markov model according to the click stream of the similar user aggregates obtained by the user relationship matrix. The content recommendation method is individualized information recommendation technology on the network, and by the method, the interesting commodity and information are recommended to the user according to the characteristics of the interest, the behavior and the personal information. The interesting information and commodity are recommended to the user in a vast database, so that the browsing time is reduced, the problems of less user rating items and more sparse items in the collaborative recommendation are solved and the accuracy of the recommendation is improved.
Description
Technical field
The present invention relates to the personalized recommendation technical field, recommend interested commodity and information to the user according to user's interest characteristics, behavior and personal information.Personalized recommendation is usually used in ecommerce and social pattern's network application based on the data mining of magnanimity, can reduce the time of browsing for the user recommends its information of interest and commodity in huge data.The present invention is specially a kind of based on Markov chain and combine the content recommendation method of user context information.
Background technology
The personalized recommendation technology is a technology that has huge applications to be worth.The personalized recommendation technology is constantly used by various ecommerce type websites and social pattern website in recent years, for the user provides them institute's information of interest and commodity.The personalized recommendation technology is suggested in nineteen ninety-five the earliest.After this constantly by development and application in e-commerce field, and for e-commerce website has brought huge interests, like Amazon.Many in recent years social pattern's network applications also in various degree use commending system, such as bean cotyledon, with thinking that the user recommends information of interest.
The method of personalized recommendation technology mainly comprises following three kinds:
1) based on the proposed algorithm of correlation rule;
2) content-based proposed algorithm;
3) collaborative filtering recommending algorithm.
Based on the proposed algorithm of correlation rule, at first excavate correlation rule formation rule storehouse, then for the user provides the corresponding recommended project, but its extensibility can not satisfy the demands.
Content-based proposed algorithm is the similarity between definition project and the project, then for the user recommend with its browse or the similar project of project of mistake interested.But such algorithm has very large difficulty to music during the project of very difficult extraction content such as film.And content-based proposed algorithm can only be found similar project, but can't recommend the user to have other intermediate item of interest.
The collaborative filtering recommending algorithm is in customer group, to seek similar user, and comprehensive then these similar users predict that to the evaluation of a certain project this user is to this project fancy grade.Collaborative filtering is a more welcome technology.It can be recommended such as music, film relatively more complicated project, also can guarantee the novelty of recommending simultaneously.But evaluation of user information is sometimes very sparse, possibly cause user's similarity to be inaccurate, thereby makes that the project of being recommended is not liked by the user.Simultaneously the performance of collaborative filtering recommending algorithm may be lower after user and the number of entry significantly increase.
Summary of the invention
The problem that the present invention will solve is: there is deficiency in various degree in the method for existing personalized recommendation technology, possibly interested project can not accomplish comprehensive recommendation for the user, can not overcome scalability problem and sparse property problem in the proposed algorithm.
Technical scheme of the present invention is: a kind of content recommendation method based on multi Markov chain, obtain user's clickstream data through the website, and user context information, it is analyzed, and generate the commending contents model; When a user produces new click steam, utilize the model of current clickstream data and commending contents to produce the interested project of user's possibility, and recommend the user; May further comprise the steps:
1), master pattern is set up: set up master pattern, comprise each user's Markov model, customer relationship matrix and the clustering criteria function that is used to estimate the cluster result quality;
2), the model learning stage: model is learnt, is merged similar Markov model, and utilize the click data of background similar users fill to merge the back Markov model zero row, just default information;
3), the user recommends: utilize the model of current click of user and group of living in, recommend.
Step of the present invention is specially:
1), master pattern is set up:
1.1), write down and extract each user's clickstream data, said click steam information is based on the click steam information of control;
1.2), utilize clickstream data that each user is set up Markov model, comprise transition matrix A and original state λ, the user gathers G:
Among the transition matrix A, a state of each page X representation model, X
tThe expression current state, X
T-1Then represent the state of eve, establish P
Ij=(X
t=x
j| X
T-1=x
i), 0<i<n, 0<j<n, n are total number of users, i.e. P
IjExpression is by state x
iTransfer to state x
jProbability, the user pointed as A do not click page X
tThe time, P appears
T1, P
T2... P
Tn, this delegation can't calculate, and is set to zero row,
Original state λ=(p
i)=(p
I2, p
I2P
In);
When Markov model only by user u
1Clickstream data when setting up, user's set is G={u
1;
1.3), from website registered user's register-file, obtain user context information; Comprise age of user, sex, educational background, work, region; Set up the customer relationship matrix based on these user context information, and utilize user context information to confirm the similitude between the user;
1.4), set up to estimate the clustering criteria function of cluster result quality, obtain initial criteria functional value Z;
2), the model learning stage:
2.1), calculate the similarity in twos between each transition matrix, confirm the similarity between all Markov models then;
2.2), set similarity threshold; Merge the Markov model that similarity surpasses threshold value; And the transition matrix and the original state of the Markov model after the calculating merging; The user of this moment gathers all users that G has comprised the Markov model representative that merges, and deletes merged Markov model simultaneously;
2.3), according to step 2.2) user of the Markov model representative of the merging that obtains, in step 1.3) search similar user in the customer relationship matrix that obtains, constitute set GS by similar user;
2.4), utilize similar users, promptly gather the user's of GS click steam information and fill 2.2) in zero row of transition matrix of Markov model after the merging that obtains;
2.5), calculate to merge the criterion function of back cluster: to step 2.2) in each feasible Markov model Merge Scenarios all merge, and calculation criterion functional value is selected wherein maximum criterion function value Z
1, compare with initial criteria functional value Z, if Z
1>Z then calculates the similarity between any two of the Markov model of current merging, gets back to step 2.2) carry out all feasible merging, promptly secondary merges, and chooses the criterion function value Z that maximum secondary merges
2With Z
1Relatively, if Z
2>Z
1Then get back to step 2.2) carry out three times and merge, so circulation is until the merging that obtains making the criterion function value maximum, step 2.4) Markov model of the filling that obtains finally confirms, gets into step 2.6);
2.6), study finishes;
3), utilize model to carry out user's recommendation:
3.1), the user produces new clickstream data, write down this clickstream data and be used for the study of model next time;
3.2), confirm the residing Markov model of user, comprise transition matrix and original state; If the user is new user,, utilize the similar user of background information to produce Markov model then according to the customer relationship matrix;
3.3), current clickstream data of user and corresponding Markov model are obtained the most popular recommendation, and be shown to the user;
3.4), finish.
Step 1.3) step of setting up the customer relationship matrix described in is following:
1.3.1), according to the user context information data, set up the user context information matrix;
1.3.2), calculate the similarity between two two users according to the user context information matrix;
1.3.3), set up the customer relationship matrix:
If current total k user, wherein S
IjThe similarity of expression user I and user J,
Step 2.3) the described customer relationship matrix that utilizes is sought the background similar users, and step is following:
2.3.1), obtaining step 2.2) all q of Markov model representative user of the merging that obtains;
2.3.2), for q user, in the customer relationship matrix, check that each user's institute is expert at, with the descending ordering of the similarity of being expert at, m before choosing obtains each user m maximum user of similarity separately;
2.3.3), obtain m user of q group, with obtaining similar users S set G after its merging.
Further, step 2.4) fill the zero capable of transition matrix, step is following:
2.4.1), obtain the click steam information of similar users;
2.4.2), obtain in the transition matrix of the Markov model after the merging because data lack imponderable project;
2.4.3), fill imponderable project in the transition matrix with the click steam information of similar users, i.e. zero row.
Step 3.2) utilize the affiliated Markov model of user to recommend; When the user is new user: utilize step 2.3) with 2.4) calculate active user's prediction Markov model, and recommend according to this model.
Step 3.3) obtain the most popular recommendation according to current click data and corresponding Markov model, step is following:
3.3.1), confirm that the page that the active user clicks now is X
t=x
g, and the transition matrix of residing Markov model is A
u=(p
U-ij);
3.3.2), obtain A
uIn g capable, i.e. state x
gRow, (p
U-gj)=(p
U-g1, p
U-g2... P
U-gn);
3.3.3), to all p
U-gj, 0<i<=n carries out descending sort and is p
U-gn1, p
U-gn2... P
U-gnn
3.3.4), set to recommend the top n content to be the most popular content, get top n p
U-gnj: be p
U-gn1, p
U-gn2... P
U-gnN, so the X of the corresponding page
T+1=x
N1, X
t=x
N2... X
t=x
NN, be the most popular content of being recommended.
The present invention utilizes click steam information to carry out commending contents, and to sparse problem wherein solution is provided, and just user's default information is filled through similar users.Though at first the user maybe be considerably less for the scoring of project, user's browsing data but can explicit user for the attention rate of project.Simultaneously simple number of clicks can be lost a lot of data, so utilize clickstream data to be used as the Back ground Information that the user recommends.Further contemplating light has clickstream data can not solve sparse property problem fully, utilizes user's background information to fill the sparse matrix that contains odd row once more on this basis, to improve the degree of accuracy of commending system.
Beneficial effect of the present invention:
1), than proposed algorithm based on correlation rule, the present invention has improved extensibility.Proposed algorithm based on correlation rule is having new user and data to add fashionable need the excavation again all data at every turn, and extensibility is not strong.And the present invention can be newly-built one type of the user who increases newly, utilizes the behavior of similar users to recommend for it then;
2), than content-based proposed algorithm, the present invention can provide the user outside the similar item maybe interested project;
3), compare the collaborative filtering recommending algorithm, the present invention utilizes user's click steam information to come can after different clicks, recommend Projects with Different for the user carries out project recommendation, has improved the precision of recommending, and has solved sparse property problem wherein simultaneously.
Description of drawings
Fig. 1 sets up for the master pattern among the present invention, and the schematic flow sheet in model learning stage.
Embodiment
Of the present invention based on Markov chain and combine the content recommendation method of user context information, can be applied in ecommerce and the socialization website.Can make things convenient for the user to browse web sites for the user provides interested project and link, increase the rate that places an order of e-commerce website, improve the intelligent degree of using.
Related terminological interpretation among the present invention:
1) Markov model: comprise transition matrix, original state, and the representative user gathers three parts.(A, λ G) represent with tlv triple MC.
Wherein among the transition matrix A, a state of each page X representation model, X
tThe expression current state, X
T-1Then represent the state of eve, establish P
Ij=(X
t=x
j| X
T-1=x
i), 0<i<n, 0<j<n, n are total number of users, i.e. P
IjExpression is by state x
iTransfer to state x
jProbability, when A user pointed did not click page Xt, P appearred
T1, P
T2... P
Tn, this delegation can't calculate, and is set to zero row,
Original state λ=(p
i)=(p
I2, p
I2P
In);
When Markov model only by user u
1Clickstream data when setting up, user's set is G={u
1.
2) similarity of Markov model
The dynamic perfromance of Markov model is described by transition matrix, and the similarity of model is based on the similarity of transition matrix.
To any two Markov model MC
mAnd MC
q, corresponding transition matrix A is arranged
mAnd A
q, their capable being respectively of i: (P
M-ij) and (P
Q-ij), j=1,2 ..., n, the similarity of these two row is:
Transition matrix A
mAnd A
qSimilarity do
Then the similarity of Markov model is:
MCSim (MC
m, MC
q)=2/ (Sim (A
m, A
q)+Sim (A
q, A
m)), and MCSim (MC
m, MC
q)=2/0=∞.
3) user's similarity
User's similarity is calculated as known technology, like " the collaborative filtering recommending algorithm that combines user context information ", Wu Yifan, Wang Haoran, the method for being carried in this piece paper.
4) customer relationship matrix
If total k user, wherein S
IjThe similarity of expression user I and user J.
Fig. 1 is the master pattern foundation of the present invention and the schematic flow sheet in model learning stage, may further comprise the steps:
1), master pattern establishment stage
1.1), write down and extract each user's clickstream data;
1.2), utilize data that each user is set up Markov model, comprise transition matrix and original state, establish total n user; MC
k=(A
k, λ
k, G
k), 0<k<n+1,
Wherein
G
k={ u
k;
Under the original state each user is set up a Markov model, this time each Markov model representative be the user set of having only a user.After similar in the back Markov model merged, each Markov model will be represented a plurality of users, in user's set of Markov model also correspondence comprise a plurality of users.
1.3), set up the customer relationship matrix, utilize user context information to confirm the similarity between the user:
1.4), set up to estimate the criterion function of cluster result quality.
1.5), finish;
2), the learning phase of model
2.1), calculate the similarity in twos between each transition matrix, and confirm the similarity between the different Markov models; Promptly for (A
1, A
1... A
k) in any A
iAnd A
jCalculate similarity S
Ij
2.2), set similarity threshold; Merge the Markov model that similarity surpasses threshold value; And the transfer matrix and the original state of the Markov model after the calculating merging; The user of this moment gathers all users that G has comprised the Markov model representative that merges, and deletes merged Markov model simultaneously; Here similarity threshold is set big more; Then but pooled model quantity is few more; The information that the user gathers among the G is just fewer; Zero too much row can appear; And that threshold value was set was low, and the cluster of user profile will be affected, and possibly occur that information too mixes in user's set; Can't accurately recommend, influence recommendation results by user interest.
2.3), according to step 2.2) user of the Markov model representative of the merging that obtains, in step 1.3) search similar user in the customer relationship matrix U that obtains, obtain similar users set GS (u by similar user
1, u
2... u
Gsn); It is similar on user's background information that similar users set GS gathers G with the user.When in the customer relationship matrix U, searching, such as user j, the j of inspection customer relationship matrix is capable, finds out three the highest values of this delegation, the pairing i1 that classifies as of these values, and i2, i3, the similar users of user j in relational matrix is user i1, i2, i3 so.
2.4), utilize similar users, promptly gather the user's of GS click steam information and fill 2.2) in zero row of transition matrix of Markov model after the merging that obtains.Obtain the click steam information that the user gathers G, reach new Markov model, just step 2.2 by its generation) merge the Markov model that obtains, note transition matrix wherein is A
Gij, leave out original transition matrix A simultaneously
iAnd A
jAnd their pairing original states.For matrix A
GijMiddle zero row that occurs, the clickstream data that utilizes the user to gather GS calculates to be inserted.
2.5), calculate to merge the criterion function of back cluster: to step 2.2) in each feasible Markov model Merge Scenarios all merge, and calculation criterion functional value is selected wherein maximum criterion function value Z
1, compare with initial criteria functional value Z, if Z
1>Z then calculates the similarity between any two of the Markov model of current merging, gets back to step 2.2) carry out all feasible merging, promptly secondary merges, and chooses the criterion function value Z that maximum secondary merges
2With Z
1Relatively, if Z
2>Z
1Then get back to step 2.2) carry out three times and merge, so circulation is until the merging that obtains making the criterion function value maximum, step 2.4) Markov model of the filling that obtains finally confirms, gets into step 2.6);
Have multiple choices when selecting the merging of any two Markov models according to certain similarity threshold values.Such as, ABC is three Markov models, the similarity threshold values is made as 0.7, and the similarity of A and B is 0.9, the similarity of A and C is 0.8, so just can select A and B to merge or A and C merging.Which kind of is more reasonable to need to judge these two kinds merging, and whether reasonably criterion function judges merging foundation exactly.So need in advance A and B to be merged calculation criterion functional value Sab.And then with A and C merging, calculation criterion functional value Sac.Suppose that S is the initial criteria functional value, the criterion function value when promptly merging.Judge Sab, Sac, this size of three of S.Choose the pairing scheme of maximal value.Reasonable plan that Here it is.Step 2.5) is actually and chooses reasonable plan in the feasible program.
2.6), finish.
When master pattern foundation, and accomplish when learning to form final mask, get into step (3) and utilize model to carry out user's recommendation.3), utilize model to carry out user's recommendation
3.1), the user produces new click data, writes down this click data user study of model next time.
3.2), confirm the classification at this user place, and obtain corresponding Markov model, comprise transition matrix and original state.If the user is new user, then utilize the similar user of background information to produce Markov model;
3.3), obtain the recommendation of top-N according to current click data of user and corresponding Markov model, and be shown to the user;
3.4), finish.
Step 1.1 wherein) in, record also extracts user's click steam information, and said click steam is based on the click steam of control.Based on the click steam information of control than behavior and interest that can accurate more recording user based on the click steam of the page.
Step 2.3) the described customer relationship matrix that utilizes is sought the background similar users, and step is following:
2.3.1), obtain using all n user in the classification of current Markov model;
2.3.2), for each user, in the customer relationship matrix, seek m maximum user of similarity;
2.3.3), obtain m user of n group, obtain similar users after it is merged and gather;
2.3.4), finish.
Step 3.2) Markov model that utilizes the user to belong to classification is recommended, when the user is new user: utilize 2.3) with 2.4) calculate active user's prediction Markov model, and recommend according to this model.
Step 3.3) obtain the recommendation of top-N according to current click data and corresponding Markov model, step is following:
3.3.1), confirm that the page that the active user clicks now is X
t=x
g, and the transition matrix of residing Markov model is A
u=(p
U-ij);
3.3.2), obtain A
uIn g capable, i.e. state x
gRow, (p
U-gj)=(p
U-g1, p
U-g2... P
U-gn);
3.3.3), to all p
U-gj, 0<i<=n carries out descending sort and is p
U-gn1, p
U-gn2... P
U-gnn
3.3.4), set to recommend the top n content to be the most popular content, get top n p
U-gnj: be p
U-gn1, p
U-gn2... P
U-gnN, so the X of the corresponding page
T+1=x
N1, X
t=x
N2... X
t=x
NN, be the most popular content of being recommended.
Claims (3)
1. the content recommendation method based on multi Markov chain is characterized in that obtaining user's clickstream data through the website, and user context information, it is analyzed, and generate the commending contents model; When a user produces new click steam, utilize the model of current clickstream data and commending contents to produce the interested project of user's possibility, and recommend the user; May further comprise the steps:
1), master pattern is set up: set up master pattern, comprise each user's Markov model, customer relationship matrix and the clustering criteria function that is used to estimate the cluster result quality;
2), the model learning stage: model is learnt, is merged similar Markov model, and utilize the click data of background similar users fill to merge the back Markov model zero row, just default information;
3), the user recommends: utilize the model of current click of user and group of living in, recommend;
Be specially:
1), master pattern is set up:
1.1), write down and extract each user's clickstream data, said click steam information is based on the click steam information of control;
1.2), utilize clickstream data that each user is set up Markov model, comprise transition matrix A and original state λ, the user gathers G:
Among the transition matrix A, a state of each page X representation model, X
tThe expression current state, X
T-1Then represent the state of eve, establish P
Ij=(X
t=x
j| X
T-1=x
i), 0<i<n, 0<j<n, n are total number of users, i.e. P
IjExpression is by state x
iTransfer to state x
jProbability, the user pointed as A do not click page X
tThe time, P appears
T1, P
T2... P
Tn, this delegation can't calculate, and is set to zero row,
Original state λ=(p
i)=(p
I2, p
I2P
In);
When Markov model only by user u
1Clickstream data when setting up, user's set is G={u
1;
1.3), from website registered user's register-file, obtain user context information; Comprise age of user, sex, educational background, work, region; Set up the customer relationship matrix based on these user context information, and utilize user context information to confirm the similitude between the user;
1.4), set up to estimate the clustering criteria function of cluster result quality, obtain initial criteria functional value Z;
2), the model learning stage:
2.1), calculate the similarity in twos between each transition matrix, confirm the similarity between all Markov models then;
2.2), set similarity threshold; Merge the Markov model that similarity surpasses threshold value; And the transition matrix and the original state of the Markov model after the calculating merging; The user of this moment gathers all users that G has comprised the Markov model representative that merges, and deletes merged Markov model simultaneously;
2.3), according to step 2.2) user of the Markov model representative of the merging that obtains, in step 1.3) search similar user in the customer relationship matrix that obtains, constitute set GS by similar user;
2.4), utilize similar users, promptly gather the user's of GS click steam information and fill 2.2) in zero row of transition matrix of Markov model after the merging that obtains;
2.5), calculate to merge the criterion function of back cluster: to step 2.2) in each feasible Markov model Merge Scenarios all merge, and calculation criterion functional value is selected wherein maximum criterion function value Z
1, compare with initial criteria functional value Z, if Z
1>Z then calculates the similarity between any two of the Markov model of current merging, gets back to step 2.2) carry out all feasible merging, promptly secondary merges, and chooses the criterion function value Z that maximum secondary merges
2With Z
1Relatively, if Z
2>Z
1Then get back to step 2.2) carry out three times and merge, so circulation is until the merging that obtains making the criterion function value maximum, step 2.4) Markov model of the filling that obtains finally confirms, gets into step 2.6);
2.6), study finishes;
3), utilize model to carry out user's recommendation:
3.1), the user produces new clickstream data, write down this clickstream data and be used for the study of model next time;
3.2), confirm the residing Markov model of user, comprise transition matrix and original state; If the user is new user,, utilize the similar user of background information to produce Markov model then according to the customer relationship matrix;
3.3), current clickstream data of user and corresponding Markov model are obtained the most popular recommendation, and be shown to the user;
3.4), finish;
Step 1.3) step of setting up the customer relationship matrix described in is following:
1.3.1), according to the user context information data, set up the user context information matrix;
1.3.2), calculate the similarity between two two users according to the user context information matrix;
1.3.3), set up the customer relationship matrix:
If current total k user, wherein S
IjThe similarity of expression user I and user J,
Step 2.3) the described customer relationship matrix that utilizes is sought the background similar users, and step is following:
2.3.1), obtaining step 2.2) all q of Markov model representative user of the merging that obtains;
2.3.2), for q user, in the customer relationship matrix, check that each user's institute is expert at, with the descending ordering of the similarity of being expert at, m before choosing obtains each user m maximum user of similarity separately;
2.3.3), obtain m user of q group, with obtaining similar users S set G after its merging;
Step 2.4) fill zero of transition matrix and go, step is following:
2.4.1), obtain the click steam information of similar users;
2.4.2), obtain in the transition matrix of the Markov model after the merging because data lack imponderable project;
2.4.3), fill imponderable project in the transition matrix with the click steam information of similar users, i.e. zero row.
2. a kind of content recommendation method based on multi Markov chain according to claim 1 is characterized in that step 3.2) utilize the affiliated Markov model of user to recommend; When the user is new user: utilize step 2.3) with 2.4) calculate active user's prediction Markov model, and recommend according to this model.
3. a kind of content recommendation method based on multi Markov chain according to claim 1 is characterized in that step 3.3) obtain the most popular recommendation according to current click data and corresponding Markov model, step is following:
3.3.1), confirm that the page that the active user clicks now is X
t=x
g, and the transition matrix of residing Markov model is A
u=(p
U-ij);
3.3.2), obtain A
uIn g capable, i.e. state x
gRow, (p
U-gj)=(p
U-g1, p
U-g2... P
U-gn);
3.3.3), to all p
U-gj, 0<i<=n carries out descending sort and is p
U-gn1, p
U-gn2... P
U-gnn
3.3.4), set to recommend the top n content to be the most popular content, get top n p
U-gnj: be p
U-gn1, p
U-gn2... P
U-gnN, so the X of the corresponding page
T+1=x
N1, X
t=x
N2... X
t=x
NN, be the most popular content of being recommended.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010101828444A CN101826114B (en) | 2010-05-26 | 2010-05-26 | Multi Markov chain-based content recommendation method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010101828444A CN101826114B (en) | 2010-05-26 | 2010-05-26 | Multi Markov chain-based content recommendation method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101826114A CN101826114A (en) | 2010-09-08 |
CN101826114B true CN101826114B (en) | 2012-05-09 |
Family
ID=42690034
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2010101828444A Expired - Fee Related CN101826114B (en) | 2010-05-26 | 2010-05-26 | Multi Markov chain-based content recommendation method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101826114B (en) |
Families Citing this family (33)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101658853B1 (en) | 2010-10-29 | 2016-09-22 | 삼성전자주식회사 | Method for providing phone directory service for mobile communication subscriber and therefor system |
CN102592223A (en) * | 2011-01-18 | 2012-07-18 | 卓望数码技术(深圳)有限公司 | Commodity recommending method and commodity recommending system |
EP2549423A1 (en) * | 2011-07-22 | 2013-01-23 | Axel Springer Digital TV Guide GmbH | Automatic determination of the relevance of recommendations in a social network |
CN102306178A (en) * | 2011-08-25 | 2012-01-04 | 盛乐信息技术(上海)有限公司 | Video recommendation method and device |
CN102508870B (en) * | 2011-10-10 | 2013-09-11 | 南京大学 | Individualized recommending method in combination of rating data and label data |
CN102402766B (en) * | 2011-12-27 | 2017-07-28 | 北京京东尚科信息技术有限公司 | A kind of user interest modeling method based on web page browsing |
EP2680209A1 (en) * | 2012-06-29 | 2014-01-01 | Axel Springer Digital TV Guide GmbH | Device and method for automatic generation of a recommendation |
CN102779182A (en) * | 2012-07-02 | 2012-11-14 | 吉林大学 | Collaborative filtering recommendation method for integrating preference relationship and trust relationship |
CN102779190B (en) * | 2012-07-03 | 2014-12-03 | 北京大学 | Rapid detection method for hot issues of timing sequence massive network news |
CN104123321B (en) * | 2013-04-28 | 2019-01-15 | 百度在线网络技术(北京)有限公司 | A kind of determining method and device for recommending picture |
CN104281622B (en) * | 2013-07-11 | 2017-12-05 | 华为技术有限公司 | Information recommendation method and device in a kind of social media |
CN103593770A (en) * | 2013-10-24 | 2014-02-19 | 清华大学 | Amazon elastic compute cloud bidding method based on Markov model |
CN103559252A (en) * | 2013-11-01 | 2014-02-05 | 桂林电子科技大学 | Method for recommending scenery spots probably browsed by tourists |
US9361397B2 (en) * | 2013-11-14 | 2016-06-07 | International Business Machines Corporation | Device data personalization |
CN104090900B (en) * | 2013-12-20 | 2015-07-29 | 深圳市腾讯计算机系统有限公司 | The mass data real-time computing technique that a kind of temperature associates, device and server |
US9405838B2 (en) * | 2013-12-27 | 2016-08-02 | Quixey, Inc. | Determining an active persona of a user device |
CN104899236B (en) * | 2014-11-13 | 2019-01-29 | 深圳市腾讯计算机系统有限公司 | A kind of comment information display methods, apparatus and system |
CN105590234A (en) * | 2014-12-30 | 2016-05-18 | 中国银联股份有限公司 | Method and system for recommending commodities to target users |
CN105183748B (en) * | 2015-07-13 | 2018-11-06 | 电子科技大学 | A kind of combination forecasting method based on content and scoring |
CN105824912A (en) * | 2016-03-15 | 2016-08-03 | 平安科技(深圳)有限公司 | Personalized recommending method and device based on user portrait |
CN107329887B (en) * | 2016-04-28 | 2020-08-14 | 腾讯科技(深圳)有限公司 | Data processing method and device based on recommendation system |
CN105956146A (en) * | 2016-05-12 | 2016-09-21 | 腾讯科技(深圳)有限公司 | Article information recommending method and device |
CN106339505B (en) * | 2016-09-27 | 2019-09-27 | 电子科技大学 | A kind of music recommended method based on markov chain |
CN106650800B (en) * | 2016-12-08 | 2020-06-30 | 南京航空航天大学 | Markov equivalence class model distributed learning method based on Storm |
CN106980989B (en) * | 2016-12-29 | 2021-01-19 | 中国银联股份有限公司 | Merchant recommendation method based on user behavior characteristic analysis |
CN108572981B (en) * | 2017-03-10 | 2021-03-05 | 中移(杭州)信息技术有限公司 | Information recommendation method and device |
CN107169830B (en) * | 2017-05-15 | 2020-11-03 | 南京大学 | Personalized recommendation method based on clustering PU matrix decomposition |
CN107947972B (en) * | 2017-11-16 | 2021-01-15 | 长安大学 | Detection method and detection device for sensing abnormal operation state of node |
CN110020168A (en) * | 2017-12-27 | 2019-07-16 | 艾迪普(北京)文化科技股份有限公司 | A kind of three-dimensional material recommended method based on big data |
CN108243357A (en) * | 2018-01-25 | 2018-07-03 | 北京搜狐新媒体信息技术有限公司 | A kind of video recommendation method and device |
CN109800325B (en) * | 2018-12-26 | 2021-10-26 | 北京达佳互联信息技术有限公司 | Video recommendation method and device and computer-readable storage medium |
CN111814903B (en) * | 2020-07-21 | 2022-04-01 | 上海数鸣人工智能科技有限公司 | Method for analyzing user sensitivity to marketing activities based on DPI clustering |
CN112733018B (en) * | 2020-12-31 | 2022-12-06 | 哈尔滨工程大学 | Session recommendation method based on graph neural network GNN and multi-task learning |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5640553A (en) * | 1995-09-15 | 1997-06-17 | Infonautics Corporation | Relevance normalization for documents retrieved from an information retrieval system in response to a query |
CN101320375A (en) * | 2008-07-04 | 2008-12-10 | 浙江大学 | Digital book search method based on user click action |
CN101482884A (en) * | 2009-01-21 | 2009-07-15 | 华东师范大学 | Cooperation recommending system based on user predilection grade distribution |
-
2010
- 2010-05-26 CN CN2010101828444A patent/CN101826114B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5640553A (en) * | 1995-09-15 | 1997-06-17 | Infonautics Corporation | Relevance normalization for documents retrieved from an information retrieval system in response to a query |
CN101320375A (en) * | 2008-07-04 | 2008-12-10 | 浙江大学 | Digital book search method based on user click action |
CN101482884A (en) * | 2009-01-21 | 2009-07-15 | 华东师范大学 | Cooperation recommending system based on user predilection grade distribution |
Also Published As
Publication number | Publication date |
---|---|
CN101826114A (en) | 2010-09-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101826114B (en) | Multi Markov chain-based content recommendation method | |
CN104166668B (en) | News commending system and method based on FOLFM models | |
CN104935963B (en) | A kind of video recommendation method based on timing driving | |
CN103678618B (en) | Web service recommendation method based on socializing network platform | |
CN102982042B (en) | A kind of personalization content recommendation method, platform and system | |
CN103632290B (en) | A kind of based on the mixing recommendation method recommending probability fusion | |
CN105045907A (en) | Method for constructing visual attention-label-user interest tree for personalized social image recommendation | |
CN107833117A (en) | A kind of Bayes's personalized ordering for considering label information recommends method | |
CN105183727A (en) | Method and system for recommending book | |
CN105045931A (en) | Video recommendation method and system based on Web mining | |
CN104933239A (en) | Hybrid model based personalized position information recommendation system and realization method therefor | |
CN103353880B (en) | A kind of utilization distinctiveness ratio cluster and the data digging method for associating | |
CN105426514A (en) | Personalized mobile APP recommendation method | |
CN104317900A (en) | Multiattribute collaborative filtering recommendation method oriented to social network | |
CN103729359A (en) | Method and system for recommending search terms | |
CN107391582B (en) | The information recommendation method of user preference similarity is calculated based on context ontology tree | |
CN110737834B (en) | Recommendation method and device for business objects, storage medium and computer equipment | |
CN104679835A (en) | Book recommending method based on multi-view hash | |
CN103150667B (en) | A kind of personalized recommendation method based on body construction | |
CN105138508A (en) | Preference diffusion based context recommendation system | |
CN104111941A (en) | Method and equipment for information display | |
CN105678590B (en) | Cloud model-based topN recommendation method for social network | |
CN107562947A (en) | A kind of Mobile Space-time perceives the lower dynamic method for establishing model of recommendation service immediately | |
CN107391670A (en) | A kind of mixing recommendation method for merging collaborative filtering and user property filtering | |
CN104899229A (en) | Swarm intelligence based behavior clustering system |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120509 Termination date: 20160526 |
|
CF01 | Termination of patent right due to non-payment of annual fee |