CN106354862A - Multidimensional individualized recommendation method in heterogeneous network - Google Patents
Multidimensional individualized recommendation method in heterogeneous network Download PDFInfo
- Publication number
- CN106354862A CN106354862A CN201610804696.2A CN201610804696A CN106354862A CN 106354862 A CN106354862 A CN 106354862A CN 201610804696 A CN201610804696 A CN 201610804696A CN 106354862 A CN106354862 A CN 106354862A
- Authority
- CN
- China
- Prior art keywords
- user
- project
- similarity
- matrix
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a multidimensional individualized recommendation method in a heterogeneous network which includes: 1, obtaining information; 2, establishing a similarity matrix between the user and item; 3, establishing a semi-structured heterogeneous information network; calculating the degree of correlation between users and items and between users and users under different meta paths based on the semi-structured heterogeneous information network; distributing different weights to the degree of correlation of each meta path to form a similarity matrix between users and other types of entities in the semi-structured heterogeneous information network; 4, distributing different weights and integrating the degree of correlation between users and items and between users and users with user's preference to items to form the final similarity matrix; 5, recommending a plurality of items with big similarity in the final similarity matrix to users. The multidimensional individualized recommendation method is added with user subordinate information and item subordinate information, considers the rich semantic information between users and items, and improves the recommendation accuracy and percentage of coverage.
Description
Technical field
The present invention relates to the relevant search method based on first path and combine field feedback in Heterogeneous Information network
With the multidimensional personalized recommendation method of dependency information, belong to the technical field of data mining and machine learning.
Background technology
In recent years, Heterogeneous Information network represents various different relations and can be accurate between dissimilar entity because being good at
The different context that ground is distinguished in information network is excavated more significant knowledge and is widely studied.Wherein, two realities are represented
Between body, first path of relation is a unique concept in Heterogeneous Information network, and different first paths represents different physics meanings
Justice, the abundant semantic feature that first path is contained, is a very important greatly feature of Heterogeneous Information network.Many isomery letters
In breath network data mining task based on first path carried out it is recommended that most active research direction even more therein it
One.
Mass data explosive increase today it is recommended that system as a kind of solve information overload effective means, weight
Want and essential, and personalized recommendation system is by setting up the binary crelation between user and product, existing using user
Selection course or similarity relationships excavate the potential object interested of each user, can accomplish more targeted recommended users
The real product needing.Mostly traditional commending system is to be made an issue of based on user and project matrix, and wherein most successful impact is the widest
General is collaborative filtering.Traditional collaborative filtering is divided into the collaborative filtering based on memory and the association based on model
Same filter algorithm, both only make use of the information of sole user's project matrix it is recommended that limited precision.In the face of conventional recommendation system
Single relation and do not consider item contents and user property and do not account between user and article between system user and article
The shortcomings of abundant semantic information, how will be powerful with Heterogeneous Information network for the conventional recommendation method containing field feedback
Integrated information integration capability and the abundant semantic information containing combine, and carry out personalized recommendation, be one have to be solved
Problem.
Content of the invention
For the deficiencies in the prior art, the invention provides multidimensional personalized recommendation method in a kind of heterogeneous network;
The present invention provides multidimensional personalized recommendation method in a kind of heterogeneous network based on first path.The method is using different
On the basis of the different semantic information that first path is contained, in conjunction with field feedback, provide the user a kind of the individual of various dimensions
Propertyization recommends method so that recommendation results have more accuracy, practicality and multiformity.
Term is explained
1st, Non-negative Matrix Factorization, is to find nonnegative matrix and make.Both equatioies are very difficult essentially equal in the calculation.In meter
Often rule iteration is updated according to certain in calculation and update two multipliers, the distance (as Euclidean distance) at two ends is full about above formula
The size of our settings of foot, stops iteration;
2nd, Heterogeneous Information network (heterogeneous informationnetwork), can be referred to as heterogeneous network.
If the number of the type on the type on summit or side is more than one in an information network, this information network is referred to as isomery letter
Breath network, otherwise for isomorphism information network.
The technical scheme is that
Multidimensional personalized recommendation method in a kind of heterogeneous network, concrete steps include:
(1) obtain information: obtain social relations information between user's self information and user, between project self information and project
The interactive relationship information and user feedback information to project;Described project refers to the other types in addition to user type;User
Self information, for example, the age of user, area, hobby etc.;Social relations information between user, for example, pays close attention to and is concerned pass
System, friendss etc.;Project, for example, film;When project refers to film, project self information, it may for example comprise film upper
Reflect type of time, film etc.;Interactive relationship information between project, it may for example comprise have between film project and protagonist project deduction and
The interactive relationship such as deduced, user has the interactive informations such as mark, scoring to film project.Come with film recommendation network the most classical
Say, the polytype such as sports representative's film, performer, director, movies category label.The electricity similar to it can be recommended for user
Shadow, performer, director, movies category etc., reach the purpose of multidimensional recommendation.
(2) build similarity matrix between user and project: between described user and project, similarity moment matrix representation user is to item
The direct reflection whether mesh is liked, the preference to project for the user;
(3) build semi-structured Heterogeneous Information network, based on semi-structured Heterogeneous Information network, calculate under different units path
Similarity and any user between each user and each project, obtain under this yuan of path user and user it
Between similarity matrix and user and project between;Similarity between row and row in similarity matrix between user and project
Characterize user between similarity, in the similarity matrix between user and project similarity characterization project between the column and the column it
Between similarity;A kind of user can be obtained to the similarity matrix between the user being generated under each first path and project
Between similarity and project between similarity, according to the domain knowledge weights different to different similarity distribution, melt
Be combined into the similarity matrix between user and project in semi-structured Heterogeneous Information network, the value in matrix illustrate each user with
Similarity size between each project;Value is bigger, and similarity is higher;
(4) different weights are distributed, by similarity matrix, use between user and user in semi-structured Heterogeneous Information network
Similarity matrix between family and project, is merged to the feedback information of project with user, obtains final similarity matrix;
For example, when project refers to film, user refers to the scoring to film for the user to the feedback information of project.With user
The feedback information of project is merged so that result is more personalized.
(5) some projects larger for similarity in final similarity matrix are recommended user.
According to currently preferred, described step (2), specifically refer to: if there is the score data to project for the user, obtain
Take the score data to project for the family, construct rating matrix, and adopt Algorithms of Non-Negative Matrix Factorization completion rating matrix;Otherwise,
Then calculate similarity matrix between user and project using the user's dependency information collected and project affiliation information;
Dependency information includes: social network information (trusting relationship, vermicelli relaying relationship and friendss etc.) and user's tribute
The information (information such as dynamic that the label of user, user comment content, user geographical position and user issue) offered.
For example in film recommendation network, user's dependency information includes age, sex and its label etc., when project refers to electricity
During shadow, project affiliation information includes the show time of film, film type etc.;When project instructs and drills, project affiliation packet
The type including the film at age, sex and its frequent director represents;
Using Algorithms of Non-Negative Matrix Factorization completion rating matrix, because the rating matrix dimension according to constructed by the information collected
Number is very high and be very sparse, and most straightforward approach is exactly by rating matrix completion with matrix disassembling method herein, and uses
Family is all nonnegative number to the score data of project, so select Algorithms of Non-Negative Matrix Factorization herein.
According to currently preferred, described step (2), concrete steps include:
If 1. there is the score data to project for the user, original for user rating matrix is designated as um×n, m is number of users,
N is the number of project, enters step 2., if there is not the score data to project for the user, enters step 3.;
2. to um×nTwo nonnegative matrixes p are decomposed into using Algorithms of Non-Negative Matrix Factorizationm×kAnd qk×nSo that matrix after decomposingMeet formula ():
In formula (), matrix pm×kFor user's factor matrix, matrix pm×kMiddle all elements non-negative, matrix qk×nFor project because
Submatrix, matrix qk×nMiddle all elements non-negative;
By formula () by higher dimensional matrix um×nIt is decomposed into the user factor matrix p of two low-dimensionalsm×kWith project factor square
Battle array qk×n.
3. adjustment cosine similarity or Pearson came similarity coefficient distance metric mode, calculate any user and any project it
Between similarity, constitute similarity matrix.
By original rating matrix um×nResolve into two matrix pm×kAnd qk×nProduct, this problem is converted into machine
Regression problem in study.Herein, the loss function of employing is original rating matrix um×nWith the rating matrix rebuildingBetween error square as loss function, for preventing over-fitting, regularization term can be added;
According to currently preferred, described step (3), concrete steps include:
A, the semi-structured Heterogeneous Information network g of structure, g=(v, e), v represent top in semi-structured Heterogeneous Information network g
Point, e represents the side in semi-structured Heterogeneous Information network g;Any two entity in semi-structured Heterogeneous Information network g is represented
Be and, if passing through first path between this two entities up to (two inter-entity exist and connect sides and there is certain pass in other words
System), then there is side e in this two inter-entity;Any one node v ∈ v, v belong to the element in node type set a;A=a1,
a2, a3, al+1, any one side e ∈ e, e belong to the element in the type set r on side;Described entity includes described item
Mesh, described user;
If two entity types all represent the project of same type, such as two summits all represent film item types
Or all representing protagonist item types etc., then this two summits are same types, and the entity type of mark is identical.If
Article two, side is connected to same type of source node and destination node, then two types when belonging to same.
B, the semi-structured Heterogeneous Information network g being built according to step a, extract the net of semi-structured Heterogeneous Information network g
Network pattern, is expressed as tg=(a, r);Network schemer tg=(a, r) is the basic mould of semi-structured Heterogeneous Information network g=(v, e)
Plate, is limited to the type on nodes and side, and exactly this restriction result in the semi-structured of network, directs net
The semantic meaning of network is explored.
C, the network schemer according to semi-structured Heterogeneous Information network g, build first path p: first path p is defined on network mould
Formula tgOn=(a, r), in the case of not causing ambiguity, directly represent first path p with node type, represent as formula () institute
Show:
P=(a1a2a3···al+1)(ⅱ)
For example,Directly it is expressed as apa, if p1=(a1a2a3···al), p2=
(b1b2b3···bk), then p=(p1p2)=(a1a2a3···alb2b3···bk), for example, first path mam (film-drill
Member-film) represent is by the film of same actor, and first path mgm (film-classification-film) then represents and belongs to together
Film in a classification.
Otherwise, shown in the expression of first path p such as formula ():
In formula (),Illustrate from entity a1To entity al+1Complexity pass
System, r=r1οr2οr3ο···οrl;Attended operation on ο representation relation, the length of first path p is the number of r;
Different units path represents different physical significances, and abundant semantic feature is contained in first path, is Heterogeneous Information
Network is different from a very important greatly feature of homogeneous network.Node type can directly be used in the case of not causing ambiguity
To represent first path.
D, according to different first paths, apply hetesim algorithm to obtain similar between user and project under different units paths
Degree matrix.
According to currently preferred, described step (4), comprising: similarity moment between the user obtain step (2) and project
The similarity matrix of battle array and the user that obtains of step (3) and other types inter-entity is merged, and obtains between user and project
Similarity matrix.
According to currently preferred, using the fusion method taking multiple result meansigma methodss, user that step (2) is obtained with
The similarity matrix of the user that between project, similarity matrix and step (3) obtain and other types inter-entity is merged, and obtains
Similarity matrix between user and project.The project of final user and any one type can be obtained after merging
Between similarity.
The invention has the benefit that
1st, add user's dependency information and project affiliation information in traditional recommendation method, and consider user with
Abundant semantic information between project, improves accuracy rate and the coverage rate of recommendation.
2nd, consider the interactive information between user and project in recommendation process, embody the preference of user, real
Show personalized recommendation.
3rd, the present invention can calculate the similarity between any user and any type of project, is not limited to push away to user
Recommend the commodity of a certain type, any type of commodity can be recommended to user, realize the recommendation of various dimensions.
Brief description
Fig. 1 is the scene description schematic diagram of embodiment film recommendation network.
Fig. 2 is Fig. 1 corresponding heterogeneous network network schemer schematic diagram.
Fig. 3 (a) is corresponding unit path " film-actor-movie " schematic diagram in Fig. 2.
Fig. 3 (b) is corresponding unit path " film-classification-film " schematic diagram in Fig. 2.
Fig. 3 (c) is corresponding unit path " user-film-actor-movie " schematic diagram in Fig. 2.
Fig. 3 (d) is corresponding unit path " user-film-classification-film " schematic diagram in Fig. 2.
Fig. 4 is embodiment user's dependency information, film dependency information example schematic.
Fig. 5 is hetesim algorithm computational methods schematic diagram.
Fig. 6 is the detail flowchart of the present invention.
Specific embodiment
With reference to Figure of description and embodiment, the present invention is further qualified, but not limited to this.
Embodiment
The present embodiment is with reference to film recommendation network, as shown in figure 1, film recommendation network is an isomery letter having the right oriented
, there are 5 kinds of different entity types in this network in breath network: user's (include user 1, user 2), film (include film 1,
Film 2, film 3, film 4), director's (include director 1, director 2), act the leading role (including performer 1, performer 2) and movies category (inclusion
Cinematic genre 1, cinematic genre 2).Can there is its protagonist, director and movies category entity for each movie property with it
It is connected.User gives, to the film that it has been seen, the scoring that 1--5 divides.According to the attribute information of user, according to viewing interest by user
It is divided into different user's groups, division can be overlapping, and same user's group is liked the film of identical style or likes same star to direct
Deng.In the present embodiment, it is film, interest group, director, performer that user recommends that it may like etc., realize personalization
Multidimensional is recommended.Multidimensional personalized recommendation method in this film recommendation network, concrete steps include:
The rating matrix u characterizing user preference degree is built to the score data of film according to user, in u matrix, data is non-
Negative, dimension are very high and extremely sparse.Non-negative Matrix Factorization refers to for a higher dimensional matrix to be decomposed into two or more low-dimensionals
The form of nonnegative matrix product, reaches the purpose of dimensionality reduction, and the null value in former sparse matrix can be predicted.It is applied to this reality
The user applying example, in film score data, original for user rating matrix is designated as um×n, m is number of users, and n is film number.
It is applied to Algorithms of Non-Negative Matrix Factorization and be decomposed into two nonnegative matrixes pm×kAnd qk×nSo that matrix after decomposingMeet:Matrix pm×kMiddle all elements non-negative, expression is pass between m user and k theme
System, and matrix qk×nMiddle all elements non-negative, expression is relation between k theme and n film.By original scoring square
Battle array um×nResolve into two matrix pm×kAnd qk×nProduct, this problem is converted into the regression problem in machine learning.
Herein, the loss function of employing is original rating matrix um×nWith the rating matrix rebuildingBetween
Error square as loss function, for preventing over-fitting, regularization term can be added;
For ensureing that in calculating process, all matrixes are non-negative, so using the property taken advantage of more new regulation, as follows:
And
Progressive alternate, until algorithm is finally restrained.
Using above-mentioned process, obtain matrix pm×kAnd qk×n, so just for user i, commodity j can be given a mark:
Finally give the user after process the rating matrix of project is characterized with the preference to this project for the user.This enforcement
There is the score data to film for the user in the data that example obtains, the preference matrix to film for the user can be obtained.
If there is no user's score information in the information collected, characterize user with the dependency information of user, with project
Dependency information represents project, and in the present embodiment, the dependency information situation of user and film is as shown in Figure 4.In the present embodiment, will
Conventional film types is as a sequential attribute pond.The label of the film types liked with this user is as the genus of user
Property, i-th user uiValue according to the position assignment 1 and 0 in attribute pond, then ui=(0,1,0,0,0,1,1,0,...), with this
Direct the attribute as director for the type of the film often led, j-th director djValue right in attribute pond according to whether existing
The attribute answering position is in corresponding position assignment 1 and 0, then dj=(1,0,0,0,0,1,1,1,...).It is used herein as cosine similar
Degree is as the similarity between measuring similarity criterion calculation user and director.Arbitrarily there is one between user and any director
Similarity value, constitutes the similarity matrix ud between user and director, can characterize the preference to director for the user.With
Similarity matrix ua between family and performer, similarity matrix ug between user and film types build principle and process with upper
Identical.
Obtain: the preference matrix to film, director, performer and film types for the user.Information architecture according to obtaining has
Weigh oriented semi-structured Heterogeneous Information network, information network can be defined as directed graph g=(v, e), any one node v
∈ v broadly falls into the element in specific node type set a;Any one type set r when e ∈ e broadly falls into specific
In element.If two type when belonging to same, this two sides are connected to same type of source node and mesh
Node.If the number of types on the number of types of node or side is more than 1, this information network is called Heterogeneous Information network, no
It is then isomorphism information network.Next we build the network schemer of this network according to the Heterogeneous Information network information, can help
We more fully understand the type of node in a complicated Heterogeneous Information network and the type on side.
According to the definition of Heterogeneous Information network, Fig. 1 scene graph is mapped in Heterogeneous Information network for we, according to heterogeneous network
The definition of network directed graph, is mapped as the corresponding node of in figure article entity, and such as film, director, performer etc. is an entity,
With the type of entity as respective nodes type.If being related between any two entity, between corresponding two nodes
There is even side.
Network schemer tg=(a, r) is the foundation formss of heterogeneous network g=(v, e), the type to nodes and side
Limited, exactly this restriction result in the semi-structured of network, the semantic meaning directing network is explored.The present embodiment
Middle network schemer, as shown in Fig. 2 can form user's group by interaction between user and user, is commented between user and film
The relation divided and be scored, has, between film and director, the relation directed and direct, and has deduction and quilt between film and performer
Relation between deduction, has the relation belonging to and comprising between film and classification.In order to preferably analyze in Heterogeneous Information network
The implicit various different meaning of one's words, can analyze the first path in network, and first path is a series of between two object types
Relational sequence, abundant semantic information and the knowledge hidden can effectively be excavated in network by first path.First path p is
It is defined on network schemer tgOn=(a, r), such asIllustrate from entity a1To entity
al+1Complexity relation, p=r1οr2οr3ο···οrl.Wherein, the attended operation on ο representation relation.The length of first path p
The number of relation r of being.First path p=directly can also be represented with node type in the case of not causing ambiguity
(a1a2a3···al+1), for example:Apa can be directly expressed as.If p1=(a1a2a3···
al), p2=(b1b2b3···bk), then p=(p1p2)=(a1a2a3···alb2b3···bk).Different units path represents
Different physical significances, the abundant semantic feature that first path is contained, is that Heterogeneous Information network is different from homogeneous network
A very important greatly feature.Fig. 3 (a) is first path " film-actor-movie " schematic diagram.Fig. 3 (b) is first path " electricity
Shadow-classification-film " schematic diagram.Fig. 3 (c) is first path " user-film-actor-movie " schematic diagram.Fig. 3 (d) is first path
" user-film-classification-film " schematic diagram.Represent in first path mam (film-actor-movie) is by same performer master
The film drilled, and first path mgm (film-classification-film) then represents the film belonging to a classification.
The different semantic information exemplary plot that table 1 is contained by different first paths:
Table 1
First path | Physical significance |
Film-actor-movie (mam) | Film by same actor |
User-film-actor-movie (umam) | The film by same actor that user has seen |
Film-classification-film (mgm) | Belong to the film of a classification |
User-film-classification-film (umgm) | What user had seen belongs to the film of a classification |
After the completion of building first path step, build entity under different units path followed by application hetesim algorithm (of the same race
Type and different types) between degree of association, form correlation matrix.The basic thought of hetesim is: if two entity quilts
Two similar entities point to or are respectively directed to two similar entities, then this two entities are also similar.For example: similar grinds
The person of studying carefully can deliver the paper of similar topic, and similar client can buy similar commodity etc..Given introductory path p=r1οr2οr3
ο···οrl, the definition of the degree of association hetesim between any two entity s and t is:
o(s|r1) it is that s is based on relation r1Out-degree, i (t | rl) it is that entity t is based on relation rlIn-degree.If s is not any
Out-degree or t do not have any in-degree, then the similarity defining two inter-entity is 0.
As shown in figure 5, calculating the degree of association based on first path apc between author (a) tom and meeting (c) kdd, then:
Because o (tom | ap)={ p1,p2And i (kdd | pc)={ p1,p2, so hetesim (tom, dd | apc)=
0.5.That is along this yuan of path, the probability that tom and kdd meets is 0.5.Hetesim can calculate any type of two
The degree of association based on different units path for the individual entity.In the present embodiment, each user and arbitrarily electricity are calculated using hetesim algorithm
Degree of association between shadow, constitutes correlation matrix hete_um, user and director, performer, the film types between user and film
Between correlation matrix hete_ud, hete_ua, hete_ug can also obtain.
In above-mentioned recommendation method, correlation matrix between the user by the user preference obtaining degree matrix and obtaining and project
Merged, finally given the similarity matrix between user and project.Fusion method is a lot, present invention employs better simply
Take the fusion method of multiple result meansigma methodss, the project of final user and any one type can be obtained after merging
Between similarity.For example, user a recommends film, directly takes that data line of user a in family-film similarity matrix
If the maximum cadre's film of middle Similarity value.
Claims (6)
1. in a kind of heterogeneous network multidimensional personalized recommendation method it is characterised in that concrete steps include:
(1) obtain information: obtain social relations information between user's self information and user, interactive between project self information and project
The relation information and user feedback information to project, described project refers to the other types in addition to user type;
(2) build similarity matrix between user and project: between described user and project, similarity moment matrix representation user likes to project
The direct reflection whether liked, the preference to project for the user;
(3) build semi-structured Heterogeneous Information network, based on semi-structured Heterogeneous Information network, calculate each under different units path
Similarity and any user between individual user and each project, obtain under this yuan of path between user and user and
Similarity matrix between user and project;Similarity characterization between row and row in similarity matrix between user and project
Similarity between user, in the similarity matrix between user and project between similarity characterization project between the column and the column
Similarity;Similarity matrix between the user being generated under each first path and project can be obtained between a kind of user
Similarity and project between similarity, according to the domain knowledge weights different to different similarity distribution, be fused to
Similarity matrix between user and project in semi-structured Heterogeneous Information network, the value in matrix illustrate each user and each
Similarity size between project;
(4) different weights are distributed, by similarity matrix, Yong Huyu between user and user in semi-structured Heterogeneous Information network
Similarity matrix between project, is merged to the feedback information of project with user, obtains final similarity matrix;
(5) some projects larger for similarity in final similarity matrix are recommended user.
2. in a kind of heterogeneous network according to claim 1 multidimensional personalized recommendation method it is characterised in that described step
(2), specifically refer to: if there is the score data to project for the user, obtain the score data to project for the user, construction scoring square
Battle array, and adopt Algorithms of Non-Negative Matrix Factorization completion rating matrix;Otherwise, then using the user's dependency information collected and project from
Genus information calculates similarity matrix between user and project.
3. in a kind of heterogeneous network according to claim 2 multidimensional personalized recommendation method it is characterised in that described step
(2), concrete steps include:
If 1. there is the score data to project for the user, original for user rating matrix is designated as um×n, m is number of users, and n is
2. the number of project, enter step, if there is not the score data to project for the user, enters step 3.;
2. to um×nTwo nonnegative matrixes p are decomposed into using Algorithms of Non-Negative Matrix Factorizationm×kAnd qk×nSo that matrix after decomposing
Meet formula ():
In formula (), matrix pm×kFor user's factor matrix, matrix pm×kMiddle all elements non-negative, matrix qk×nFor project factor square
Battle array, matrix qk×nMiddle all elements non-negative;
3. adjustment cosine similarity or Pearson came similarity coefficient distance metric mode, calculates between any user and any project
Similarity, constitutes similarity matrix.
4. in a kind of heterogeneous network according to claim 1 multidimensional personalized recommendation method it is characterised in that described step
(3), concrete steps include:
A, the semi-structured Heterogeneous Information network g of structure, g=(v, e), v represent summit in semi-structured Heterogeneous Information network g, e table
Show the side in semi-structured Heterogeneous Information network g;Any two entity in semi-structured Heterogeneous Information network g is expressed as and,
If passing through first path between this two entities up to this two inter-entity have side e;Any one node v ∈ v, v belong to
Element in node type set a;A=a1, a2, a3, al+1, any one side e ∈ e, e belong to the type set r on side
In element;Described entity includes described project, described user;
B, the semi-structured Heterogeneous Information network g being built according to step a, extract the network mould of semi-structured Heterogeneous Information network g
Formula, is expressed as tg=(a, r);
C, the network schemer according to semi-structured Heterogeneous Information network g, build first path p: first path p is defined on network schemer tg
On=(a, r), in the case of not causing ambiguity, directly represent first path p with node type, represent as shown in formula ():
P=(a1a2a3···al+1) (ⅱ)
Otherwise, shown in the expression of first path p such as formula ():
In formula (),Illustrate from entity a1To entity al+1Complexity relation, Attended operation on representation relation, the length of first path p is the number of r;
D, according to different first paths, apply hetesim algorithm to obtain the similarity moment between user and project under different units paths
Battle array.
5. in a kind of heterogeneous network according to claim 1 multidimensional personalized recommendation method it is characterised in that include: will
User in the semi-structured Heterogeneous Information network that between user that step (2) obtains and project, similarity matrix and step (3) obtain
Similarity matrix and between project is merged, and obtains final similarity matrix.
6. in a kind of heterogeneous network according to claim 1 multidimensional personalized recommendation method it is characterised in that many using taking
The fusion method of individual result meansigma methodss, half that between the user that obtain step (2) and project, similarity matrix and step (3) obtain
In structuring Heterogeneous Information network, the similarity matrix between user and project is merged, and obtains final similarity matrix.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610804696.2A CN106354862A (en) | 2016-09-06 | 2016-09-06 | Multidimensional individualized recommendation method in heterogeneous network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610804696.2A CN106354862A (en) | 2016-09-06 | 2016-09-06 | Multidimensional individualized recommendation method in heterogeneous network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106354862A true CN106354862A (en) | 2017-01-25 |
Family
ID=57859094
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610804696.2A Pending CN106354862A (en) | 2016-09-06 | 2016-09-06 | Multidimensional individualized recommendation method in heterogeneous network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106354862A (en) |
Cited By (29)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107562795A (en) * | 2017-08-01 | 2018-01-09 | 广州市香港科大***研究院 | Recommendation method and device based on Heterogeneous Information network |
CN107577710A (en) * | 2017-08-01 | 2018-01-12 | 广州市香港科大***研究院 | Recommendation method and device based on Heterogeneous Information network |
CN107747947A (en) * | 2017-10-23 | 2018-03-02 | 电子科技大学 | A kind of collaboration itinerary based on user's history GPS track recommends method |
CN107944629A (en) * | 2017-11-30 | 2018-04-20 | 北京邮电大学 | A kind of recommendation method and device based on heterogeneous information network representation |
CN108304496A (en) * | 2018-01-11 | 2018-07-20 | 上海交通大学 | Node similarity relation detection method based on composite unit path in Heterogeneous Information net |
CN108763515A (en) * | 2018-05-31 | 2018-11-06 | 天津理工大学 | A kind of time-sensitive personalized recommendation method decomposed based on probability matrix |
CN108805628A (en) * | 2018-06-19 | 2018-11-13 | 东北师范大学 | A kind of electronic commerce recommending method |
CN109102111A (en) * | 2018-07-26 | 2018-12-28 | 北京工商大学 | A method of measurement director cooperates possibility with performer |
CN109271582A (en) * | 2018-08-20 | 2019-01-25 | 东南大学 | It is a kind of based on the recommendation method for personalized information with attribute member path |
CN109635201A (en) * | 2018-12-18 | 2019-04-16 | 苏州大学 | The heterogeneous cross-platform association user account method for digging of social networks |
CN109685630A (en) * | 2019-01-09 | 2019-04-26 | 苏州大学 | E-commerce purchases by group recommended method and system |
CN109710835A (en) * | 2018-11-15 | 2019-05-03 | 中国人民解放军国防科技大学 | Heterogeneous information network recommendation method with time weight |
CN110555050A (en) * | 2018-03-30 | 2019-12-10 | 华东师范大学 | heterogeneous network node representation learning method based on meta-path |
CN110580196A (en) * | 2019-09-12 | 2019-12-17 | 北京邮电大学 | Multi-task reinforcement learning method for realizing parallel task scheduling |
CN111191056A (en) * | 2020-01-02 | 2020-05-22 | 中国联合网络通信集团有限公司 | Multimedia recommendation method and device |
CN111241380A (en) * | 2018-11-28 | 2020-06-05 | 富士通株式会社 | Method and apparatus for generating recommendations |
CN111476322A (en) * | 2020-05-22 | 2020-07-31 | 哈尔滨工程大学 | Meta-path mining method based on feature optimization |
CN111507788A (en) * | 2019-01-31 | 2020-08-07 | 阿里巴巴集团控股有限公司 | Data recommendation method and device, storage medium and processor |
CN112148931A (en) * | 2020-09-29 | 2020-12-29 | 河北工业大学 | Meta path learning method for high-order abnormal picture classification |
CN112221159A (en) * | 2020-10-21 | 2021-01-15 | 腾讯科技(深圳)有限公司 | Virtual item recommendation method and device and computer readable storage medium |
CN112434086A (en) * | 2020-12-04 | 2021-03-02 | 崔秀芬 | Information flow mining method based on cloud computing and big data and cloud computing interaction center |
CN112433655A (en) * | 2020-12-04 | 2021-03-02 | 崔秀芬 | Information flow interaction processing method based on cloud computing and cloud computing verification interaction center |
CN112532755A (en) * | 2021-02-18 | 2021-03-19 | 广州汇图计算机信息技术有限公司 | Interest list pushing system based on heterogeneous information network |
CN112650944A (en) * | 2020-12-29 | 2021-04-13 | 哈尔滨理工大学 | Collaborative filtering recommendation algorithm based on high-order connection graph |
CN112733039A (en) * | 2021-01-27 | 2021-04-30 | 中国科学院地理科学与资源研究所 | Personalized recommendation method and system and electronic equipment |
CN112836119A (en) * | 2021-01-26 | 2021-05-25 | 西安理工大学 | Scenic spot recommendation method based on weighted heterogeneous information network |
CN113268629A (en) * | 2021-04-29 | 2021-08-17 | 西安交通大学 | Heterogeneous picture singing list multi-label recommendation method fusing node preference |
CN113836395A (en) * | 2021-08-30 | 2021-12-24 | 河南大学 | Heterogeneous information network-based service developer on-demand recommendation method and system |
CN113935804A (en) * | 2021-10-15 | 2022-01-14 | 燕山大学 | Semantic recommendation method based on reinforcement learning and weighted element path |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104077351A (en) * | 2014-05-26 | 2014-10-01 | 东北师范大学 | Heterogeneous information network based content providing method and system |
CN104199826A (en) * | 2014-07-24 | 2014-12-10 | 北京大学 | Heterogeneous media similarity calculation method and retrieval method based on correlation analysis |
CN104317900A (en) * | 2014-10-24 | 2015-01-28 | 重庆邮电大学 | Multiattribute collaborative filtering recommendation method oriented to social network |
CN104850632A (en) * | 2015-05-22 | 2015-08-19 | 东北师范大学 | Generic similarity calculation method and system based on heterogeneous information network |
-
2016
- 2016-09-06 CN CN201610804696.2A patent/CN106354862A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104077351A (en) * | 2014-05-26 | 2014-10-01 | 东北师范大学 | Heterogeneous information network based content providing method and system |
CN104199826A (en) * | 2014-07-24 | 2014-12-10 | 北京大学 | Heterogeneous media similarity calculation method and retrieval method based on correlation analysis |
CN104317900A (en) * | 2014-10-24 | 2015-01-28 | 重庆邮电大学 | Multiattribute collaborative filtering recommendation method oriented to social network |
CN104850632A (en) * | 2015-05-22 | 2015-08-19 | 东北师范大学 | Generic similarity calculation method and system based on heterogeneous information network |
Cited By (47)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107577710A (en) * | 2017-08-01 | 2018-01-12 | 广州市香港科大***研究院 | Recommendation method and device based on Heterogeneous Information network |
CN107562795A (en) * | 2017-08-01 | 2018-01-09 | 广州市香港科大***研究院 | Recommendation method and device based on Heterogeneous Information network |
CN107577710B (en) * | 2017-08-01 | 2020-06-19 | 广州市香港科大***研究院 | Recommendation method and device based on heterogeneous information network |
CN107747947B (en) * | 2017-10-23 | 2021-04-30 | 电子科技大学 | Collaborative travel route recommendation method based on historical GPS (global positioning system) track of user |
CN107747947A (en) * | 2017-10-23 | 2018-03-02 | 电子科技大学 | A kind of collaboration itinerary based on user's history GPS track recommends method |
CN107944629A (en) * | 2017-11-30 | 2018-04-20 | 北京邮电大学 | A kind of recommendation method and device based on heterogeneous information network representation |
CN107944629B (en) * | 2017-11-30 | 2020-08-07 | 北京邮电大学 | Recommendation method and device based on heterogeneous information network representation |
CN108304496A (en) * | 2018-01-11 | 2018-07-20 | 上海交通大学 | Node similarity relation detection method based on composite unit path in Heterogeneous Information net |
CN110555050B (en) * | 2018-03-30 | 2023-03-31 | 华东师范大学 | Heterogeneous network node representation learning method based on meta-path |
CN110555050A (en) * | 2018-03-30 | 2019-12-10 | 华东师范大学 | heterogeneous network node representation learning method based on meta-path |
CN108763515B (en) * | 2018-05-31 | 2021-12-17 | 天津理工大学 | Time-sensitive personalized recommendation method based on probability matrix decomposition |
CN108763515A (en) * | 2018-05-31 | 2018-11-06 | 天津理工大学 | A kind of time-sensitive personalized recommendation method decomposed based on probability matrix |
CN108805628B (en) * | 2018-06-19 | 2021-03-16 | 东北师范大学 | Electronic commerce recommendation method |
CN108805628A (en) * | 2018-06-19 | 2018-11-13 | 东北师范大学 | A kind of electronic commerce recommending method |
CN109102111A (en) * | 2018-07-26 | 2018-12-28 | 北京工商大学 | A method of measurement director cooperates possibility with performer |
CN109271582A (en) * | 2018-08-20 | 2019-01-25 | 东南大学 | It is a kind of based on the recommendation method for personalized information with attribute member path |
CN109710835A (en) * | 2018-11-15 | 2019-05-03 | 中国人民解放军国防科技大学 | Heterogeneous information network recommendation method with time weight |
CN111241380A (en) * | 2018-11-28 | 2020-06-05 | 富士通株式会社 | Method and apparatus for generating recommendations |
CN111241380B (en) * | 2018-11-28 | 2023-10-03 | 富士通株式会社 | Method and apparatus for generating recommendations |
CN109635201A (en) * | 2018-12-18 | 2019-04-16 | 苏州大学 | The heterogeneous cross-platform association user account method for digging of social networks |
CN109635201B (en) * | 2018-12-18 | 2020-07-31 | 苏州大学 | Heterogeneous social network cross-platform associated user account mining method |
CN109685630B (en) * | 2019-01-09 | 2020-10-27 | 苏州大学 | E-commerce group purchase recommendation method and system |
CN109685630A (en) * | 2019-01-09 | 2019-04-26 | 苏州大学 | E-commerce purchases by group recommended method and system |
CN111507788A (en) * | 2019-01-31 | 2020-08-07 | 阿里巴巴集团控股有限公司 | Data recommendation method and device, storage medium and processor |
CN111507788B (en) * | 2019-01-31 | 2023-07-14 | 阿里巴巴华北技术有限公司 | Data recommendation method, device, storage medium and processor |
CN110580196A (en) * | 2019-09-12 | 2019-12-17 | 北京邮电大学 | Multi-task reinforcement learning method for realizing parallel task scheduling |
CN110580196B (en) * | 2019-09-12 | 2021-04-06 | 北京邮电大学 | Multi-task reinforcement learning method for realizing parallel task scheduling |
CN111191056A (en) * | 2020-01-02 | 2020-05-22 | 中国联合网络通信集团有限公司 | Multimedia recommendation method and device |
CN111476322A (en) * | 2020-05-22 | 2020-07-31 | 哈尔滨工程大学 | Meta-path mining method based on feature optimization |
CN111476322B (en) * | 2020-05-22 | 2022-05-20 | 哈尔滨工程大学 | Meta-path mining method based on feature optimization |
CN112148931A (en) * | 2020-09-29 | 2020-12-29 | 河北工业大学 | Meta path learning method for high-order abnormal picture classification |
CN112148931B (en) * | 2020-09-29 | 2022-11-04 | 河北工业大学 | Meta path learning method for high-order abnormal picture classification |
CN112221159B (en) * | 2020-10-21 | 2022-04-08 | 腾讯科技(深圳)有限公司 | Virtual item recommendation method and device and computer readable storage medium |
CN112221159A (en) * | 2020-10-21 | 2021-01-15 | 腾讯科技(深圳)有限公司 | Virtual item recommendation method and device and computer readable storage medium |
CN112434086B (en) * | 2020-12-04 | 2021-09-21 | 上海东方财富证券投资咨询有限公司 | Information flow mining method based on cloud computing and big data and cloud computing interaction center |
CN112433655A (en) * | 2020-12-04 | 2021-03-02 | 崔秀芬 | Information flow interaction processing method based on cloud computing and cloud computing verification interaction center |
CN112434086A (en) * | 2020-12-04 | 2021-03-02 | 崔秀芬 | Information flow mining method based on cloud computing and big data and cloud computing interaction center |
CN112650944A (en) * | 2020-12-29 | 2021-04-13 | 哈尔滨理工大学 | Collaborative filtering recommendation algorithm based on high-order connection graph |
CN112836119A (en) * | 2021-01-26 | 2021-05-25 | 西安理工大学 | Scenic spot recommendation method based on weighted heterogeneous information network |
CN112733039A (en) * | 2021-01-27 | 2021-04-30 | 中国科学院地理科学与资源研究所 | Personalized recommendation method and system and electronic equipment |
CN112532755A (en) * | 2021-02-18 | 2021-03-19 | 广州汇图计算机信息技术有限公司 | Interest list pushing system based on heterogeneous information network |
CN112532755B (en) * | 2021-02-18 | 2021-05-04 | 广州汇图计算机信息技术有限公司 | Interest list pushing system based on heterogeneous information network |
CN113268629A (en) * | 2021-04-29 | 2021-08-17 | 西安交通大学 | Heterogeneous picture singing list multi-label recommendation method fusing node preference |
CN113268629B (en) * | 2021-04-29 | 2023-01-03 | 西安交通大学 | Heterogeneous picture singing list multi-label recommendation method fusing node preference |
CN113836395A (en) * | 2021-08-30 | 2021-12-24 | 河南大学 | Heterogeneous information network-based service developer on-demand recommendation method and system |
CN113935804A (en) * | 2021-10-15 | 2022-01-14 | 燕山大学 | Semantic recommendation method based on reinforcement learning and weighted element path |
CN113935804B (en) * | 2021-10-15 | 2024-05-14 | 燕山大学 | Semantic recommendation method based on reinforcement learning and weighting element path |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106354862A (en) | Multidimensional individualized recommendation method in heterogeneous network | |
Sun et al. | Mining heterogeneous information networks: principles and methodologies | |
CN106802956A (en) | A kind of film based on weighting Heterogeneous Information network recommends method | |
CN103325061B (en) | A kind of community discovery method and system | |
CN103955535A (en) | Individualized recommending method and system based on element path | |
CN106250545A (en) | A kind of multimedia recommendation method and system searching for content based on user | |
CN111259133A (en) | Personalized recommendation method integrating multiple information | |
Ji et al. | Improving matrix approximation for recommendation via a clustering-based reconstructive method | |
CN111259263A (en) | Article recommendation method and device, computer equipment and storage medium | |
CN109190030A (en) | Implicit feedback recommendation method fusing node2vec and deep neural network | |
CN109034960A (en) | A method of more inferred from attributes based on user node insertion | |
CN102495837B (en) | Training method and system for digital information recommending and forecasting model | |
Avrachenkov et al. | Cooperative game theory approaches for network partitioning | |
CN109272390A (en) | The personalized recommendation method of fusion scoring and label information | |
Li et al. | Overlap community detection using spectral algorithm based on node convergence degree | |
Li et al. | Social recommendation based on trust and influence in SNS environments | |
Das et al. | Personalized movie recommendation system using twitter data | |
Irvanizam et al. | An application for smartphone preference using TODIM decision making method | |
Serrano | Intelligent recommender system for big data applications based on the random neural network | |
Mohamed et al. | Sparsity and cold start recommendation system challenges solved by hybrid feedback | |
Yin et al. | A survey of learning-based methods for cold-start, social recommendation, and data sparsity in e-commerce recommendation systems | |
CN107194492A (en) | The optimization method that a kind of businessman based on position social networks is recommended | |
Lu et al. | Social network alignment: a bi-layer graph attention neural networks based method | |
Ma et al. | The SOM based improved k-means clustering collaborative filtering algorithm in TV recommendation system | |
CN115600642B (en) | Stream media-oriented decentralization federation learning method based on neighbor trust aggregation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | 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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170125 |