CN109522481A - A kind of recommended method for expanding the user visual field based on Markov model - Google Patents
A kind of recommended method for expanding the user visual field based on Markov model Download PDFInfo
- Publication number
- CN109522481A CN109522481A CN201811347438.1A CN201811347438A CN109522481A CN 109522481 A CN109522481 A CN 109522481A CN 201811347438 A CN201811347438 A CN 201811347438A CN 109522481 A CN109522481 A CN 109522481A
- Authority
- CN
- China
- Prior art keywords
- topic
- user
- state
- interest
- preference
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 19
- 230000000007 visual effect Effects 0.000 title claims abstract description 17
- 230000007704 transition Effects 0.000 claims abstract description 20
- 239000011159 matrix material Substances 0.000 claims abstract description 17
- 238000011161 development Methods 0.000 claims abstract description 15
- 230000001394 metastastic effect Effects 0.000 claims abstract description 7
- 206010061289 metastatic neoplasm Diseases 0.000 claims abstract description 7
- 238000012546 transfer Methods 0.000 claims description 34
- 230000018109 developmental process Effects 0.000 claims description 12
- 238000013507 mapping Methods 0.000 claims description 4
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 230000036301 sexual development Effects 0.000 claims description 3
- 230000006399 behavior Effects 0.000 abstract description 4
- 238000009412 basement excavation Methods 0.000 abstract description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 208000001491 myopia Diseases 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
Abstract
Recommended method provided by the invention that expand the user visual field based on Markov model, user interest development track model carry out mathematical modeling by first-order Markov model, obtain branch mode and metastatic rule between the interested topic state of user;The generation method of topic state transition probability matrix efficiently can find the interested next topic of its most probable for user;It is gradually according to the actual situation its interest topic of user development that the method that topic is expanded, which can allow recommender system,.Present invention introduces Markov model and pass through the excavation to group of subscribers user behaviors log, the interest development track of user is abstracted, establish simple and effective user interest prediction model, under the background that existing recommender system algorithm can solve the recommended requirements under current interest topic for user well, user is helped to find other possible interested topics.
Description
Technical field
The present invention relates to recommender system fields, expand user based on Markov model more particularly, to a kind of
The recommended method in the visual field.
Background technique
Along with the high speed development of internet industry, its major application field such as: film music, video, is done shopping, is asked
Community, short-sighted frequency, news etc. are answered, all brings the information of magnanimity to user.With being skyrocketed through for data volume, people are entered
The epoch of " information overload (Information overload) ", user will find oneself interested information from bulk information
More and more difficult, search engine can allow user to find the information that they need according to keyword, but when user can not be accurate
The demand of oneself is described, search engine is with regard to helpless.In order to solve " information overload " and " demand is indefinite " the two
Problem, recommender system are come into being.Recommender system can be by analyzing the historical behavior of user, and according to the personal information of user
With the other information of institute recommendation, user is helped to find their interested contents.
Now common recommender system algorithm includes CF, CBF, there are also the recommendations that some pairs of algorithms are mixed.Wherein, CF because
It is simple and highly effective for model, it is the proposed algorithm being most widely used now.CF includes ItemCF with UserCF again,
ItemCF recommends article similar with the article liked before it to user, and article similitude therein passes through between user
Historical behavior is calculated, and two articles are liked simultaneously by more people, the two articles are more similar.And UserCF is to use
The article liked with other users similar in its interest is recommended at family, i.e., similar users is found in user's dimension, for any two
A user, if the number of articles that the two users like jointly is more, the two users are more similar.
One good recommender system, should can not only according to the preference that user has shown come for its recommended user it is interested
Content, moreover it is possible to help user's expanded field of vision, excavate out the current unknown interest of user.However, existing recommender system algorithm inclines
Xiang Yuwei user recommends similar content, this is easy that user is made to be fed up with, and ItemCF can only be according to the existing history row of user
To obtain its preference, and recommendation and the most like other articles of its preference, user's expanded field of vision can not be helped.UserCF is to use
The article that family recommends other users similar with its interest to like, can help user's expanded field of vision to a certain extent, but
For UserCF when number of users is very big, the time complexity for calculating user's similarity matrix is excessively high, and whole cost is excessively high.It is existing
Some recommender systems are helping the level in the user development visual field to take up an official post so without preferable solution, and a common method is
The flow for dividing sub-fraction is used to do random content recommendation, for example, divide 5% flow recommend at random it is unrelated with user preference
Content to user, and some feedback informations according to user on random recommendation, the preference information of Lai Gengxin user, but
It is random to recommend to be that user be given to recommend completely unrelated content without guidance and insecure, reduce user experience, lowered and recommend
The accuracy of system.
Summary of the invention
Problem is narrowed for the visual field generally existing in existing recommender system or can only be recommended most close with user preferences
Other things problems, the present invention proposes a kind of recommended method for expanding the user visual field based on Markov model, this hair
It is bright the technical solution adopted is that:
A kind of recommended method for expanding the user visual field based on Markov model, comprising the following steps:
S10. mathematical modeling is carried out by first-order Markov model, establishes user interest development track model, is used
Branch mode and metastatic rule between the interested topic state in family;
S20. historical user's preferred contents data are read, and topic is abstracted as by the content map into topic space
State point in model;
S30. the transfer number and transition probability of each topic state in topic space are calculated and generates state transfer generally
Rate matrix;
S40. a preference topic sequence being in chronological sequence sequentially generated is generated for each user, which includes all
User has shown the topic of preference;
S50. by user preference topic sequence, it is emerging that the following most probable sense of user is found from state transition probability matrix
The topic of interest, i.e. the legal transfer state of maximum probability;
S60. the interested topic of the following most probable of user found for step S50, recommending according to a certain percentage should
Content under topic.
Preferably, the branch mode and metastatic rule between the interested topic state of user are obtained in the S10
Include the following steps:
S101. it is described using interest development track of the Directed Graph Model to user, all properties collections are as follows:
I={ i1, i2..., iN, | I |=N
All topic set are as follows:
S={ s1, s2..., sM, | S |=M
For any one content i, there is a topic s belonging to iti, it is clear that there is M≤N, and has
S102. be defined to the transfer of user interest: user is current to feel when showing preference to a certain content i
Entitled s if interesti;Later, user is to another topic sjUnder a certain content j when showing preference, if current interest
Entitled sj, this just represents the transfer of a user interest, is construed as the development of a user interest;
S103. topic set S is corresponded to the vertex set V, E of digraph G is side collection, for each side, indicates primary
The transfer of user interest, DEFor the set of the connection number of each edge, PEIt is the state transition probability matrix of M × M size;
Therefore, user's sexual development locus model can be with formalized description are as follows:
G=(V, E, DE, PE)
S104. according to the historical data of user, according to the sequencing of time, the content i that it is liked for each is looked for
To the corresponding topic s of content ii, corresponding vertex v is generated on figure Gi, when the corresponding topic of the content of the user preferences generates
Variation is sjWhen, corresponding vertex is vj, generate a line eijConnect viAnd vj, indicate the transfer of an interest state, while dij
It is automatic to add 1, dijInitial value be 0, aforesaid operations, d are carried out for the historical data of all usersijIt indicates from viIt is transferred to
vjTotal degree.
Preferably, the S30 includes the following steps:
S301. it calculates from state viIt is transferred to state vjTransition probability pij:
Wherein, dijIt is topic state from viIt is transferred to vjTransfer number, diFor state viIt is transferred to other stateful
The sum of transfer number;
S302. according to calculated result, transition probability matrix P is generatedE:
Preferably, the S40 step includes the following steps:
S401. all users set are as follows:
U={ u1, u2..., uK, | U |=K, wherein K indicates the serial number of user;
S402. to any user ui, it reads the history preferred contents of the user and its preferred contents is subjected to topic mapping,
The preference topic sequence for being directed to the user is generated according to chronological order tnFor the time
Node, all topics under the sequence do not repeat, if having repeated a certain topic in user's history preference topic, take the words
Position of the timing node occurred earliest as the topic in the sequence is inscribed, is recommended according to the preference topic sequence of the user for it
The corresponding content of topic in the sequence.
Preferably, the step S50 is specific as follows:
S501: system is that user finds new topic every corresponding time γ, and γ is adjustable parameter;
S502. family u is preferentially takeniNewest interested topicPass through state transition probability matrix PE, findMost
Maximum probabilityNext topic state of transferWhereinIt can formalization representation are as follows:
If S503.Then by probability value successively decrease sequential selection transfer next topic state
If S504. passing through topicLegal next topic state can not be found, then fromIn take topicInto
Jargon topic extension, is arranged an adjustable parameter α here, indicates that maximum can expand number, if finding looking into for next topic state
It looks for number to be greater than α and then stops topic expansion, ifIn all topics when can not all find legal next topic state,
Stop topic expanding.
Preferably, the step S60 is specific as follows:
One proportion adjustable parameter beta is set, when finding the interested topic of most probableWhen, recommender system is with ratio beta
Recommend topic for userUnder content;When user session is inscribedUnder table of contents reveal the feedback of " interested " after, will
TopicIt is added to the preference topic sequence of the userIn.
Compared with prior art, the beneficial effect of technical solution of the present invention is:
Present invention introduces Markov model and by the excavation to group of subscribers user behaviors log, the interest of user is developed
Track is abstracted, and is established simple and effective user interest prediction model, can is well in existing recommender system algorithm
User solves under the background of the recommended requirements under current interest topic, and user is helped to find other possible interested topics.
Detailed description of the invention
Fig. 1 is the flow chart of the recommended method provided by the invention that expand the user visual field based on Markov model.
Fig. 2 is the user interest development track model of the recommended method for expanding the user visual field based on Markov model
Schematic diagram.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, only for illustration, Bu Nengli
Solution is the limitation to this patent.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative labor
Every other embodiment obtained under the premise of dynamic, shall fall within the protection scope of the present invention.
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
Embodiment 1
Shown in Fig. 1~2, a kind of recommended method for expanding the user visual field based on Markov model, including following step
It is rapid:
S10. mathematical modeling is carried out by first-order Markov model, establishes user interest development track model, is used
Branch mode and metastatic rule between the interested topic state in family;
S20. historical user's preferred contents data are read, and topic is abstracted as by the content map into topic space
State point in model;
S30. the transfer number and transition probability of each topic state in topic space are calculated and generates state transfer generally
Rate matrix;
S40. a preference topic sequence being in chronological sequence sequentially generated is generated for each user, which includes all
User has shown the topic of preference;
S50. by user preference topic sequence, it is emerging that the following most probable sense of user is found from state transition probability matrix
The topic of interest, i.e. the legal transfer state of maximum probability;
S60. the interested topic of the following most probable of user found for step S50, recommending according to a certain percentage should
Content under topic.
Preferably, the branch mode and metastatic rule between the interested topic state of user are obtained in the S10
Include the following steps:
S101. it is described using interest development track of the Directed Graph Model to user, all properties collections are as follows:
I={ i1, i2..., iN, | I |=N
All topic set are as follows:
S={ s1, s2..., sM, | s |=M
For any one content i, there is a topic s belonging to iti, it is clear that there is M≤N, and has
S102. be defined to the transfer of user interest: user is current to feel when showing preference to a certain content i
Entitled s if interesti;Later, user is to another topic sjUnder a certain content j when showing preference, if current interest
Entitled sj, this just represents the transfer of a user interest, is construed as the development of a user interest;
S103. topic set S is corresponded to the vertex set V, E of digraph G is side collection, for each side, indicates primary
The transfer of user interest, DEFor the set of the connection number of each edge, PEIt is the state transition probability matrix of M × M size;
Therefore, user's sexual development locus model can be with formalized description are as follows:
G=(V, E, DE, PE)
S104. according to the historical data of user, according to the sequencing of time, the content i that it is liked for each is looked for
To the corresponding topic s of content ii, corresponding vertex v is generated on figure Gi, when the corresponding topic of the content of the user preferences generates
Variation is sjWhen, corresponding vertex is vj, generate a line eijConnect viAnd vj, indicate the transfer of an interest state, while dij
It is automatic to add 1, dijInitial value be 0, aforesaid operations, d are carried out for the historical data of all usersijIt indicates from viIt is transferred to
vjTotal degree.
Wherein scheme, the S30 include the following steps: as a further preference
S301. it calculates from state viIt is transferred to state vjTransition probability pij:
Wherein, dijIt is topic state from viIt is transferred to vjTransfer number, diFor state viIt is transferred to other stateful
The sum of transfer number;
S302. according to calculated result, transition probability matrix P is generatedE:
Wherein scheme, the S40 step include the following steps: as a further preference
S401. all users set are as follows:
U={ u1, u2..., uK, | U |=K, wherein K indicates the serial number of user;
S402. to any user ui, it reads the history preferred contents of the user and its preferred contents is subjected to topic mapping,
The preference topic sequence for being directed to the user is generated according to chronological order tnFor the time
Node, all topics under the sequence do not repeat, if having repeated a certain topic in user's history preference topic, take the words
Position of the timing node occurred earliest as the topic in the sequence is inscribed, is recommended according to the preference topic sequence of the user for it
The corresponding content of topic in the sequence.
Wherein scheme, the step S50 are specific as follows as a further preference:
S501: system is that user finds new topic every corresponding time γ, and γ is adjustable parameter;
S502. family u is preferentially takeniNewest interested topicPass through state transition probability matrix PE, findMost
Maximum probabilityNext topic state of transferWhereinIt can formalization representation are as follows:
If S503.Then by probability value successively decrease sequential selection transfer next topic state
If S504. passing through topicLegal next topic state can not be found, then fromIn take topicInto
Jargon topic extension, is arranged an adjustable parameter α here, indicates that maximum can expand number, if finding looking into for next topic state
It looks for number to be greater than α and then stops topic expansion, ifIn all topics when can not all find legal next topic state,
Stop topic expanding.
Embodiment 2
The pseudocode of the present embodiment is as follows:
Input:
User preference content data set Dataset
System parameter α, β, γ
User u
Output:
New topic
Recommendation list under new topic
The mapping of // topic
for item∈Dataset
si=ItemToTopic (item)
S append si
// model established according to data set
G=ModelConstruct (S, Dataset)
The transfer number and transition probability of each topic state of // calculating
for i∈M
for j∈M
dij=CalculateTransferTimes (G) // calculating state transfer number
pij=CalculateTransferProbability (G) // calculating state transition probability
D append dij
P append pij
// generate state transition probability matrix
PE=ConstructMatrix (D, P)
// it is one preference topic sequence of each user maintenance
for ui∈U
For // system every time γ, searching number α with maximum is that user u finds new topic
// recommendation under new topic is generated for user with ratio beta
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair
The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description
To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this
Made any modifications, equivalent replacements, and improvements etc., should be included in the claims in the present invention within the spirit and principle of invention
Protection scope within.
Claims (6)
1. a kind of recommended method for expanding the user visual field based on Markov model, which comprises the following steps:
S10. mathematical modeling is carried out by first-order Markov model, establishes user interest development track model, obtain user's sense
Branch mode and metastatic rule between the topic state of interest;
S20. historical user's preferred contents data are read, and by the content map into topic space, topic is abstracted as model
In state point;
S30. the transfer number and transition probability of each topic state in topic space are calculated and generates state transition probability square
Battle array;
S40. a preference topic sequence being in chronological sequence sequentially generated is generated for each user, which includes all users
The topic of preference is shown;
S50. by user preference topic sequence, it is interested that the following most probable of user is found from state transition probability matrix
Topic, i.e. the legal transfer state of maximum probability;
S60. the interested topic of the following most probable of user found for step S50, recommends the topic according to a certain percentage
Under content.
2. the recommended method according to claim 1 that expand the user visual field based on Markov model, feature exist
In the branch mode and metastatic rule obtained between the interested topic state of user in the S10 includes the following steps:
S101. it is described using interest development track of the Directed Graph Model to user, all properties collections are as follows:
I={ i1, i2..., iN, | I |=N
All topic set are as follows:
S={ s1, s2..., sM, | S |=M
For any one content i, there is a topic s belonging to iti, it is clear that there is M≤N, and has
S102. be defined to the transfer of user interest: user is when showing preference to a certain content i, current interest
If entitled si;Later, user is to another topic sjUnder a certain content j when showing preference, it is entitled if current interest
sj, this just represents the transfer of a user interest, is construed as the development of a user interest;
S103. topic set S is corresponded to the vertex set V, E of digraph G is that side collection indicates a user for each side
The transfer of interest, DEFor the set of the connection number of each edge, PEIt is the state transition probability matrix of M × M size;Therefore,
User's sexual development locus model can be with formalized description are as follows:
G=(V, E, DE, PE)
S104. according to the historical data of user, according to the sequencing of time, the content i that it is liked for each finds this
The corresponding topic s of content ii, corresponding vertex v is generated on figure Gi, when the corresponding topic of the content of the user preferences generates variation
For sjWhen, corresponding vertex is vj, generate a line eijConnect viAnd vj, indicate the transfer of an interest state, while dijAutomatically
Add 1, dijInitial value be 0, aforesaid operations, d are carried out for the historical data of all usersijIt indicates from viIt is transferred to vj's
Total degree.
3. the recommended method according to claim 2 that expand the user visual field based on Markov model, feature exist
In the S30 includes the following steps:
S301. it calculates from state viIt is transferred to state vjTransition probability pij:
Wherein, dijIt is topic state from viIt is transferred to vjTransfer number, diFor state viIt is transferred to other stateful transfers
The sum of number;
S302. according to calculated result, transition probability matrix P is generatedE:
4. the recommended method according to claim 3 that expand the user visual field based on Markov model, feature exist
In the S40 step includes the following steps:
S401. all users set are as follows:
U={ u1, u2..., uK, | U |=K, wherein K indicates the serial number of user;
S402. to any user ui, it reads the history preferred contents of the user and its preferred contents is subjected to topic mapping, according to
Chronological order generates the preference topic sequence for being directed to the user tnFor timing node,
All topics under the sequence do not repeat, if having repeated a certain topic in user's history preference topic, take the topic most
The sequence is recommended according to the preference topic sequence of the user for it in position of the timing node early occurred as the topic in the sequence
The corresponding content of topic in column.
5. the recommended method according to claim 4 that expand the user visual field based on Markov model, feature exist
In the step S50 is specific as follows:
S501: system is that user finds new topic every corresponding time γ, and γ is adjustable parameter;
S502. family u is preferentially takeniNewest interested topicPass through state transition probability matrix PE, findMost probably
RateNext topic state of transferWhereinIt can formalization representation are as follows:
If S503.Then by probability value successively decrease sequential selection transfer next topic state
If S504. passing through topicLegal next topic state can not be found, then fromIn take topicIt is talked about
Topic extension, is arranged an adjustable parameter α here, indicates that maximum can expand number, if finding the lookup number of next topic state
Then stop topic expansion greater than α, ifIn all topics when can not all find legal next topic state, also stop
Topic is expanded.
6. the recommended method according to claim 5 that expand the user visual field based on Markov model, feature exist
In the step S60 is specific as follows:
One proportion adjustable parameter beta is set, when finding the interested topic of most probableWhen, recommender system is user with ratio beta
Recommend topicUnder content;When user session is inscribedUnder table of contents reveal the feedback of " interested " after, by topicIt is added to the preference topic sequence of the userIn.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2018113204217 | 2018-11-07 | ||
CN201811320421 | 2018-11-07 |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109522481A true CN109522481A (en) | 2019-03-26 |
Family
ID=65776590
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811347438.1A Pending CN109522481A (en) | 2018-11-07 | 2018-11-13 | A kind of recommended method for expanding the user visual field based on Markov model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109522481A (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102646114A (en) * | 2012-02-17 | 2012-08-22 | 清华大学 | News topic timeline abstract generating method based on breakthrough point |
CN108388674A (en) * | 2018-03-26 | 2018-08-10 | 百度在线网络技术(北京)有限公司 | Method and apparatus for pushed information |
-
2018
- 2018-11-13 CN CN201811347438.1A patent/CN109522481A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102646114A (en) * | 2012-02-17 | 2012-08-22 | 清华大学 | News topic timeline abstract generating method based on breakthrough point |
CN108388674A (en) * | 2018-03-26 | 2018-08-10 | 百度在线网络技术(北京)有限公司 | Method and apparatus for pushed information |
Non-Patent Citations (1)
Title |
---|
李科霖: "一种融合话题和行为的在线问答社区领域专家发现方法", 《计算机与现代化》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5432243B2 (en) | Send and react to media object queries | |
US11122009B2 (en) | Systems and methods for identifying geographic locations of social media content collected over social networks | |
US20130304818A1 (en) | Systems and methods for discovery of related terms for social media content collection over social networks | |
US20130297581A1 (en) | Systems and methods for customized filtering and analysis of social media content collected over social networks | |
US20130297694A1 (en) | Systems and methods for interactive presentation and analysis of social media content collection over social networks | |
US8255421B2 (en) | Communication assistance device, communication assistance method, and communication assistance program | |
CN109522420B (en) | Method and system for acquiring learning demand | |
US11061974B2 (en) | Facilitating discovery of information items using dynamic knowledge graph | |
US20150032492A1 (en) | Methods of Identifying Relevant Content and Subject Matter Expertise for Online Communities | |
US20150032751A1 (en) | Methods and Systems for Utilizing Subject Matter Experts in an Online Community | |
US20110307469A1 (en) | System and method for query suggestion based on real-time content stream | |
US10311072B2 (en) | System and method for metadata transfer among search entities | |
CN104239440A (en) | Method and device for representing search results | |
CN103200279A (en) | Recommending method and cloud server | |
CN110188248A (en) | Data processing method, device and electronic equipment based on news question and answer interactive system | |
WO2020073526A1 (en) | Trust network-based push method, apparatus, computer device, and storage medium | |
CN112966091A (en) | Knowledge graph recommendation system fusing entity information and heat | |
CN116244513A (en) | Random group POI recommendation method, system, equipment and storage medium | |
CN110008396B (en) | Object information pushing method, device, equipment and computer readable storage medium | |
CN109299368B (en) | Method and system for intelligent and personalized recommendation of environmental information resources AI | |
CN109522481A (en) | A kind of recommended method for expanding the user visual field based on Markov model | |
Fernández et al. | Let's go to the cinema! A movie recommender system for ephemeral groups of users | |
CN112333085B (en) | Social method and electronic device | |
Liu et al. | Automatic suggestion of presentation image for storytelling | |
CN113641897A (en) | Recommendation method and device based on session text, electronic equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190326 |
|
RJ01 | Rejection of invention patent application after publication |