CN112765488A - Recommendation method, system and equipment fusing social network and knowledge graph - Google Patents
Recommendation method, system and equipment fusing social network and knowledge graph Download PDFInfo
- Publication number
- CN112765488A CN112765488A CN202110099046.3A CN202110099046A CN112765488A CN 112765488 A CN112765488 A CN 112765488A CN 202110099046 A CN202110099046 A CN 202110099046A CN 112765488 A CN112765488 A CN 112765488A
- Authority
- CN
- China
- Prior art keywords
- similarity
- user
- users
- calculating
- social
- 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 35
- 239000011159 matrix material Substances 0.000 claims abstract description 68
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 18
- 230000004927 fusion Effects 0.000 claims description 61
- 230000006399 behavior Effects 0.000 claims description 27
- 239000013598 vector Substances 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 18
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 238000011478 gradient descent method Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000011273 social behavior Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
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
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Evolutionary Biology (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Animal Behavior & Ethology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application discloses a recommendation method, a system and equipment fusing a social network and a knowledge graph, wherein influence factors of users and articles are mined from the social network and the knowledge graph respectively, similar users and similar articles are calculated, and then the similar users and the similar articles are fused into a traditional matrix decomposition recommendation model to form a new recommendation model, and the technical problem that the recommendation is influenced by the data sparsity and the cold start problem existing in the conventional recommendation system is solved.
Description
Technical Field
The application relates to the technical field of network social information recommendation, in particular to a recommendation method, system and device fusing a social network and a knowledge graph.
Background
When social network sites are flooded, people can add friends through the network, increase attention and become fans, and the social behaviors form a huge social network. In order to make good use of social network resources and achieve resource sharing and pushing, a recommendation system is adopted to push information related to user interests to users. The recommendation system mainly utilizes the behavior information of the user on the project to dig out the personalized requirements of the user, and actively provides the information meeting the requirements of the user through the interest model of the user, so that the recommendation system becomes an important research field for providing personalized services for the user and is widely applied. However, the recommendation performance is affected by the data sparsity and the cold start problem existing in the existing recommendation system.
Disclosure of Invention
The application provides a recommendation method, a recommendation system and recommendation equipment fusing a social network and a knowledge graph, which are used for solving the technical problem that the recommendation is influenced by the problems of data sparsity and cold start existing in the conventional recommendation system.
In view of the above, a first aspect of the present application provides a recommendation method for fusing a social network and a knowledge graph, including:
acquiring a social network of a user, and calculating social similarity among the users in the social network;
calculating the trust between users according to the scoring matrix and the social network;
calculating the score similarity based on the user and the project similarity based on the behavior according to the score matrix;
calculating the semantic similarity of the project according to the knowledge graph;
calculating project fusion similarity according to the project semantic similarity and the behavior-based project similarity;
calculating user fusion similarity according to the scoring similarity based on the users, the trust between the users and the social similarity between the users;
merging the project fusion similarity and the user fusion similarity into a basic matrix decomposition model to obtain a recommendation model fusing a social network and a knowledge graph;
and minimizing the recommendation model fusing the social network and the knowledge graph according to a gradient descent method to obtain the prediction score of the user on the unscored target item.
Optionally, the obtaining a social network of the user and calculating a social similarity between the users in the social network include:
acquiring a social network of a user;
training a social network of a user by using a graph convolution neural network, and learning node embedded expression of the user with social relation in a low-dimensional vector space;
and calculating the social similarity among the users by adopting a first cosine similarity function.
Optionally, the calculating the inter-user trust according to the scoring matrix and the social network includes:
calculating user project weights according to the scoring matrix and the social network, and calculating the trust level among users according to the user project weights, wherein the calculation formula of the trust level among users is as follows:
wherein, Wu,vWeight of edge for user U to user v, f (U)u,Uv) A union of items that are commonly scored for user u and user v.
Optionally, the calculating the score similarity based on the user and the item similarity based on the behavior according to the score matrix includes:
according to the scoring matrix, a second cosine similarity function is adopted to calculate and obtain the scoring similarity based on the user, wherein the second cosine similarity function is as follows:
wherein, UiFor the ith user, UjFor the jth user, RikAs a scoring matrix Rm×n(ii) the value of the rating of the ith user to the kth project, RjkAs a scoring matrix Rm×nThe scoring value of the jth user to the kth project, m is the number of users, and n is the number of projects;
and calculating to obtain the project similarity based on the behaviors by adopting a third cosine similarity function according to the scoring matrix.
Optionally, the calculating the semantic similarity of the project according to the knowledge graph includes:
performing vectorization representation on entities and relations in the knowledge graph by using a TransE algorithm on the basis of retaining semantic information to obtain a feature matrix of a project;
and calculating the semantic similarity of the project by adopting a fourth cosine similarity function according to the feature matrix of the project.
Optionally, the calculating a project fusion similarity according to the project semantic similarity and the behavior-based project similarity includes:
and fusing the item semantic similarity and the item similarity based on the behavior by adopting a fusion factor theta belonging to [0, 1] to obtain the item fusion similarity.
Optionally, the calculating a user fusion similarity according to the score similarity based on the users, the trust between the users, and the social similarity between the users includes:
fusing the scoring similarity based on the users, the trust between the users and the social similarity between the users by using a fusion factor alpha belonging to [0, 1], beta belonging to [0, 1], gamma belonging to [0, 1] to obtain the user fusion similarity, wherein the user fusion similarity is as follows:
wherein α + β + γ is 1, simsocial(Ui,Uj) In order to provide social similarity between users,for inter-user confidence, simscore(Ui,Uj) Is based on the similarity of scores of users.
Optionally, the recommendation model fusing the social network and the knowledge graph is as follows:
wherein, Ii,jThe scoring state of the user i on the item j is 1 if the user i scores the item j, otherwise, the user i scores the item j, and the scoring state is 0, lambda1、λ2、λ3、λ4The method is to avoid the over-fitting regular term generated in the parameter learning process of the model, wherein F represents F norm, RijIn order to predict the score(s),representation and article VjA similar set of k neighboring items,representation and user UiA similar set of k neighbor users.
In a second aspect, the present application provides a recommendation system fusing a social network and a knowledge graph, including:
the device comprises an acquisition unit, a judgment unit and a processing unit, wherein the acquisition unit is used for acquiring a social network of a user and calculating the social similarity among the users in the social network;
the trust degree unit is used for calculating the trust degree between the users according to the scoring matrix and the social network;
the first calculation unit is used for calculating the score similarity based on the user and the project similarity based on the behavior according to the score matrix;
the second calculation unit is used for calculating the semantic similarity of the project according to the knowledge graph;
the third calculation unit is used for calculating the item fusion similarity according to the item semantic similarity and the item similarity based on the behaviors;
the first fusion unit is used for calculating user fusion similarity according to the scoring similarity based on the users, the trust between the users and the social similarity between the users;
the second fusion unit is used for fusing the project fusion similarity and the user fusion similarity into a basic matrix decomposition model to obtain a recommendation model fusing the social network and the knowledge graph;
and the prediction unit is used for minimizing the recommendation model fusing the social network and the knowledge graph according to a gradient descent method to obtain the prediction score of the user on the unscored target item.
A third aspect of the present application provides a recommendation device fusing social networks and knowledge-graphs, the device comprising a processor and a memory:
the memory is used for storing program codes and transmitting the program codes to the processor;
the processor is configured to execute any of the methods for recommending a converged social network and knowledge graph according to the first aspect according to instructions in the program code.
According to the technical scheme, the embodiment of the application has the following advantages:
the application provides a recommendation method fusing a social network and a knowledge graph, which comprises the following steps: acquiring a social network of a user, and calculating social similarity among the users in the social network; calculating the trust between users according to the scoring matrix and the social network; calculating the score similarity based on the user and the project similarity based on the behavior according to the score matrix; calculating the semantic similarity of the project according to the knowledge graph; calculating project fusion similarity according to the project semantic similarity and the behavior-based project similarity; calculating user fusion similarity according to the scoring similarity based on the users, the trust between the users and the social similarity between the users; merging the project fusion similarity and the user fusion similarity into a basic matrix decomposition model to obtain a recommendation model fusing a social network and a knowledge graph; and minimizing the recommendation model fusing the social network and the knowledge graph according to a gradient descent method to obtain the prediction score of the user on the unscored target item. According to the method, the influence factors of the users and the articles are mined from the social network and the knowledge graph respectively, the similar users and the similar articles are calculated, and then the similar users and the similar articles are merged into the traditional matrix decomposition recommendation model to form a new recommendation model, and the technical problem that the recommendation is influenced by the problems of data sparsity and cold start existing in the conventional recommendation system is solved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other related drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic flowchart of a recommendation method for fusing a social network and a knowledge graph according to an embodiment of the present application;
FIG. 2 is a schematic diagram of the propagation process of the neural network of FIG. 2;
fig. 3 is a schematic diagram of a user relationship provided in the embodiment of the present application.
Detailed Description
In order to make the technical solutions of the present application better understood, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Example 1
For ease of understanding, referring to fig. 1, the present application provides an embodiment of a recommendation method that merges social networks and knowledge-graphs, comprising:
step 101, obtaining a social network of a user, and calculating social similarity among the users in the social network.
It should be noted that, in the embodiment of the present application, a graph convolution neural network may be used to train a social relationship network of a user, and learn node embedding expression of the user with social relationship in a low-dimensional vector space, as shown in fig. 2, fig. 2 is a schematic diagram of a propagation process of the graph convolution neural network, and in the graph convolution neural network, each layer of a hidden layer may use a propagation rule of the following formula to aggregate information, so as to form features of a next layer:
where σ (-) is the activation function,for a contiguous matrix with self-connection, INIs a matrix of the units,is composed ofThe diagonal matrix of (a) is,is a pair ofSymmetrical normalization is performed to prevent gradient disappearance or explosion, H(l)Is the output of the l-th layer, H(0)X is the user feature matrix, W(l)Is the weight matrix of the l-th layer.
An n x d user feature matrix is obtained through the above formula training, each row vector represents each user, and therefore the first cosine similarity function can be adopted to calculate the user UiAnd UjThe first cosine similarity function is expressed as:
wherein, UidAnd UjdAre respectively a user UiAnd UjThe d-th dimension vector value of (1).
And 102, calculating the trust between users according to the scoring matrix and the social network.
In the relationship graph G ═ U, E, W, where U is a set of users and E is a weight of an edge, and where W is a set of edges, and users U and v have a social relationship, an edge E may be connected between a node U and a node vu,vMeanwhile, if the user u and the user v score the same project, the edge E is obtainedu,vWeight W ofu,vThen 1 is added, and all other items are traversed to obtain the final edge weight Wu,v. Therefore, the confidence level Tu,vCan be obtained from the following equation:
wherein, f (U)u,Uv) The union of the items is scored for user u and user v together.
For example, as shown in fig. 3, user a and user B have scored the same 10 items, and their scored union number of items is 100, then the confidence level between user a and user B is:
and 103, calculating the score similarity based on the user and the project similarity based on the behavior according to the score matrix.
It should be noted that the recommendation system includes m users U ═ U (U)1,U2,…,Um) And n items V ═ V (V)1,V2,…,Vn) Then the user's rating information for the item can be represented as an m n matrix Rm×n:
Then user U may be assignediRepresented as an n-dimensional vector whose value in each dimension is the value of the corresponding user's rating for the item: u shapei=(Ri1,Ri2,…,Rin). MiningAnd obtaining the score similarity based on the user by using a second cosine similarity formula:
in the same way, the item V isiExpressed as a vector of m dimensions, whose value in each dimension is the value of the corresponding user's score for it: vi=(R1i,R2i,…,Rmi) And calculating the item similarity based on the behavior by adopting a third cosine similarity formula:
and 104, calculating the semantic similarity of the project according to the knowledge graph.
It should be noted that, on the basis of retaining semantic information, a transform algorithm may be used to perform vectorization representation on entities and relations in the knowledge graph to obtain a feature matrix of a project, and then a fourth cosine similarity function is used to calculate the semantic similarity of the project based on the knowledge graph:
and 105, calculating the item fusion similarity according to the item semantic similarity and the item similarity based on the behaviors.
It should be noted that, a fusion factor θ ∈ [0, 1] may be used to fuse the item semantic similarity and the behavior-based item similarity to obtain an item fusion similarity:
sim(Vi,Vj)=θsimkg(Vi,Vj)+(1-θ)simscore(Vi,Vj)。
and 106, calculating the user fusion similarity according to the score similarity based on the users, the trust between the users and the social similarity between the users.
It should be noted that, the fusion factor α ∈ [0, 1], β ∈ [0, 1], γ ∈ [0, 1] can be used to fuse the score similarity, the inter-user trust degree, and the inter-user social similarity based on the user, so as to obtain the user fusion similarity, where the user fusion similarity is:
wherein α + β + γ is 1, simsocial(Ui,Uj) In order to provide social similarity between users,for inter-user confidence, simscore(Ui,Uj) Is based on the similarity of scores of users.
And 107, merging the project fusion similarity and the user fusion similarity into a basic matrix decomposition model to obtain a recommendation model fusing the social network and the knowledge graph.
And 108, minimizing and fusing a recommendation model of the social network and the knowledge graph according to a gradient descent method to obtain the prediction score of the user on the unscored target item.
It should be noted that, the user similarity and the project similarity are used to merge the potential information of similar users and projects into the basic matrix decomposition model to form a unified KGSCN recommendation model, and the KGSCN algorithm considers that the feature vector representations of similar users and projects after matrix decomposition are approximate, and if the user 1 and the user 2 are similar, their feature vectors are also similar. The method comprises the following steps of minimizing a recommendation model fusing a social network and a knowledge graph by a gradient descent method, and calculating the characteristics of users and items, so as to calculate the prediction score of a target user on unscored items:
where L is the prediction score, the first term on the right of the equation is the basis of the decomposition matrix model, and secondThe item and the third item are regular items of user and item feature vectors to prevent over-fitting, the fourth item and the fifth item respectively represent similar user fusion similarity and item fusion similarity, so that the feature vector similarity between a user with high similarity and an item is as large as possible, and Ii,jThe scoring state of the user i on the item j is 1 if the user i scores the item j, otherwise, the user i scores the item j, and the scoring state is 0, lambda1、λ2、λ3、λ4The method is to avoid the over-fitting regular term generated in the parameter learning process of the model, wherein F represents F norm, namely the square sum re-evolution of corresponding elements in a matrix, and RijIn order to predict the score(s),representation and article VjA similar set of k neighboring items,representation and user UiA similar set of k neighbor users.
According to the method provided by the embodiment of the application, the influence factors of the users and the articles are mined from the social network and the knowledge graph respectively, the similar users and the similar articles are calculated, and then the similar users and the similar articles are merged into the traditional matrix decomposition recommendation model to form a new recommendation model, and the technical problem that the recommendation is influenced by the problems of data sparsity and cold start existing in the conventional recommendation system is solved.
Example 2
The application provides an embodiment of a recommendation system fusing a social network and a knowledge graph, which comprises the following steps:
the trust degree unit is used for calculating the trust degree between the users according to the scoring matrix and the social network;
the first calculation unit is used for calculating the score similarity based on the user and the project similarity based on the behavior according to the score matrix;
the second calculation unit is used for calculating the semantic similarity of the project according to the knowledge graph;
the third calculation unit is used for calculating the item fusion similarity according to the item semantic similarity and the item similarity based on the behaviors;
the first fusion unit is used for calculating user fusion similarity according to the scoring similarity based on the users, the trust between the users and the social similarity between the users;
the second fusion unit is used for fusing the project fusion similarity and the user fusion similarity into a basic matrix decomposition model to obtain a recommendation model fusing the social network and the knowledge graph;
and the prediction unit is used for minimizing the recommendation model fusing the social network and the knowledge graph according to a gradient descent method to obtain the prediction score of the user on the unscored target item.
In the embodiment of the present application, a graph convolution neural network may be used to train a social relationship network of a user, learn node embedding expression of the user with social relationship in a low-dimensional vector space, as shown in fig. 2, fig. 2 is a schematic diagram of a propagation process of the graph convolution neural network, and in the graph convolution neural network, each layer of a hidden layer may use a propagation rule of the following formula to aggregate information, so as to form a feature of a next layer:
where σ (-) is the activation function,for a contiguous matrix with self-connection, INIs a matrix of the units,is composed ofThe diagonal matrix of (a) is,is a pair ofSymmetrical normalization is performed to prevent gradient disappearance or explosion, H(l)Is the output of the l-th layer, H(0)X is the user feature matrix, W(l)Is the weight matrix of the l-th layer.
An n x d user feature matrix is obtained through the above formula training, each row vector represents each user, and therefore the first cosine similarity function can be adopted to calculate the user UiAnd UjThe first cosine similarity function is expressed as:
wherein, UidAnd UjdAre respectively a user UiAnd UjThe d-th dimension vector value of (1).
In the relationship graph G ═ U, E, W, where U is a set of users and E is a weight of an edge, and where W is a set of edges, and users U and v have a social relationship, an edge E may be connected between a node U and a node vu,vMeanwhile, if the user u and the user v score the same project, the edge E is obtainedu,vWeight W ofu,vThen 1 is added, and all other items are traversed to obtain the final edge weight Wu,v. Therefore, the confidence level Tu,vCan be obtained from the following equation:
wherein, f (U)u,Uv) The union of the items is scored for user u and user v together.
For example, as shown in fig. 3, user a and user B have scored the same 10 items, and their scored union number of items is 100, then the confidence level between user a and user B is:
it should be noted that the recommendation system includes m users U ═ U (U)1,U2,…,Um) And n items V ═ V (V)1,V2,…,Vn) Then the user's rating information for the item can be represented as an m n matrix Rm×n:
Then user U may be assignediRepresented as an n-dimensional vector whose value in each dimension is the value of the corresponding user's rating for the item: u shapei=(Ri1,Ri2,…,Rin). And obtaining the score similarity based on the user by adopting a second cosine similarity formula:
in the same way, the item V isiExpressed as a vector of m dimensions, whose value in each dimension is the value of the corresponding user's score for it: vi=(R1i,R2i,…,Rmi) And calculating the item similarity based on the behavior by adopting a third cosine similarity formula:
it should be noted that, on the basis of retaining semantic information, a transform algorithm may be used to perform vectorization representation on entities and relations in the knowledge graph to obtain a feature matrix of a project, and then a fourth cosine similarity function is used to calculate the semantic similarity of the project based on the knowledge graph:
and fusing the item semantic similarity and the behavior-based item similarity by adopting a fusion factor theta belonging to [0, 1] to obtain item fusion similarity:
sim(Vi,Vj)=θsimkg(Vi,Vj)+(1-θ)simscore(Vi,Vj)。
the fusion factor alpha belongs to [0, 1], beta belongs to [0, 1], gamma belongs to [0, 1] and is fused based on the scoring similarity of users, the trust between users and the social similarity between users to obtain the user fusion similarity, wherein the user fusion similarity is as follows:
wherein α + β + γ is 1, simsocial(Ui,Uj) In order to provide social similarity between users,for inter-user confidence, simscore(Ui,Uj) Is based on the similarity of scores of users.
And integrating potential information of similar users and items into a basic matrix decomposition model by using the user similarity and the item similarity to form a uniform KGSCN recommendation model, wherein the KGSCN algorithm considers that the feature vector representations of the similar users and the similar items after matrix decomposition are approximate, and if the users 1 and 2 are similar, the feature vectors of the similar users and the similar items are also similar. The method comprises the following steps of minimizing a recommendation model fusing a social network and a knowledge graph by a gradient descent method, and calculating the characteristics of users and items, so as to calculate the prediction score of a target user on unscored items:
wherein L is a prediction score, the first term on the right side of the equation is a basic decomposition matrix model, the second term and the third term are regularization terms of user and item feature vectors for preventing over-fitting, and the fourth term andthe fifth item represents the fusion similarity of similar users and the fusion similarity of projects respectively, so that the similarity of feature vectors between users with high similarity and projects is as large as possible, and Ii,jThe scoring state of the user i on the item j is 1 if the user i scores the item j, otherwise, the user i scores the item j, and the scoring state is 0, lambda1、λ2、λ3、λ4The method is to avoid the over-fitting regular term generated in the parameter learning process of the model, wherein F represents F norm, namely the square sum re-evolution of corresponding elements in a matrix, and RijIn order to predict the score(s),representation and article VjA similar set of k neighboring items,representation and user UiA similar set of k neighbor users.
According to the method provided by the embodiment of the application, the influence factors of the users and the articles are mined from the social network and the knowledge graph respectively, the similar users and the similar articles are calculated, and then the similar users and the similar articles are merged into the traditional matrix decomposition recommendation model to form a new recommendation model, and the technical problem that the recommendation is influenced by the problems of data sparsity and cold start existing in the conventional recommendation system is solved.
Example 3
Embodiments of a recommendation device that merges social networks and knowledge-graphs are provided herein, the device comprising a processor and a memory:
the memory is used for storing the program codes and transmitting the program codes to the processor;
the processor is configured to execute the recommendation method for fusing social networks and knowledge graphs in embodiment 1 according to instructions in the program code.
According to the device provided by the embodiment of the application, the influence factors of the users and the articles are mined from the social network and the knowledge graph respectively, the similar users and the similar articles are calculated, and then the similar users and the similar articles are merged into the traditional matrix decomposition recommendation model to form a new recommendation model, and the technical problem that the recommendation is influenced by the problems of data sparsity and cold start existing in the conventional recommendation system is solved.
The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.
Claims (10)
1. A recommendation method fusing a social network and a knowledge graph is characterized by comprising the following steps:
acquiring a social network of a user, and calculating social similarity among the users in the social network;
calculating the trust between users according to the scoring matrix and the social network;
calculating the score similarity based on the user and the project similarity based on the behavior according to the score matrix;
calculating the semantic similarity of the project according to the knowledge graph;
calculating project fusion similarity according to the project semantic similarity and the behavior-based project similarity;
calculating user fusion similarity according to the scoring similarity based on the users, the trust between the users and the social similarity between the users;
merging the project fusion similarity and the user fusion similarity into a basic matrix decomposition model to obtain a recommendation model fusing a social network and a knowledge graph;
and minimizing the recommendation model fusing the social network and the knowledge graph according to a gradient descent method to obtain the prediction score of the user on the unscored target item.
2. The recommendation method fusing social networks and knowledge graphs according to claim 1, wherein the obtaining the social networks of the users and calculating the social similarity among the users in the social networks comprises:
acquiring a social network of a user;
training a social network of a user by using a graph convolution neural network, and learning node embedded expression of the user with social relation in a low-dimensional vector space;
and calculating the social similarity among the users by adopting a first cosine similarity function.
3. The recommendation method for fusing social networks and knowledge-graphs according to claim 2, wherein the calculating the trust between users according to the scoring matrix and the social network comprises:
calculating user project weights according to the scoring matrix and the social network, and calculating the trust level among users according to the user project weights, wherein the calculation formula of the trust level among users is as follows:
wherein, Wu,vWeight of edge for user U to user v, f (U)u,Uv) A union of items that are commonly scored for user u and user v.
4. The recommendation method fusing social networks and knowledge-graphs according to claim 3, wherein the calculating the score similarity based on the user and the item similarity based on the behavior according to the score matrix comprises:
according to the scoring matrix, a second cosine similarity function is adopted to calculate and obtain the scoring similarity based on the user, wherein the second cosine similarity function is as follows:
wherein, UiFor the ith user, UjFor the jth user, RikAs a scoring matrix Rm×n(ii) the value of the rating of the ith user to the kth project, RjkAs a scoring matrix Rm×nThe scoring value of the jth user to the kth project, m is the number of users, and n is the number of projects;
and calculating to obtain the project similarity based on the behaviors by adopting a third cosine similarity function according to the scoring matrix.
5. The recommendation method fusing social networks and knowledge-graphs according to claim 4, wherein the calculating semantic similarity of items according to knowledge-graphs comprises:
performing vectorization representation on entities and relations in the knowledge graph by using a TransE algorithm on the basis of retaining semantic information to obtain a feature matrix of a project;
and calculating the semantic similarity of the project by adopting a fourth cosine similarity function according to the feature matrix of the project.
6. The method for recommending a converged social network and knowledge graph according to claim 5, wherein the calculating item-converged similarity according to the item semantic similarity and the behavior-based item similarity comprises:
and fusing the item semantic similarity and the item similarity based on the behavior by adopting a fusion factor theta belonging to [0, 1] to obtain the item fusion similarity.
7. The recommendation method for fusing social networks and knowledge-graphs according to claim 6, wherein the calculating user fusion similarity according to the user-based score similarity, the inter-user trust and the inter-user social similarity comprises:
fusing the scoring similarity based on the users, the trust between the users and the social similarity between the users by using a fusion factor alpha belonging to [0, 1], beta belonging to [0, 1], gamma belonging to [0, 1] to obtain the user fusion similarity, wherein the user fusion similarity is as follows:
8. The recommendation method for fusing social networks and knowledge-graphs according to claim 7, wherein the recommendation model for fusing social networks and knowledge-graphs is as follows:
wherein, Ii,jThe scoring state of the user i on the item j is 1 if the user i scores the item j, otherwise, the user i scores the item j, and the scoring state is 0, lambda1、λ2、λ3、λ4Avoids the over-fitting regular term generated by the model in the parameter learning process, F represents the F norm,representation and article VjA similar set of k neighboring items,representation and user UiA similar set of k neighbor users.
9. A recommendation system that merges social networks and knowledge graphs, comprising:
the device comprises an acquisition unit, a judgment unit and a processing unit, wherein the acquisition unit is used for acquiring a social network of a user and calculating the social similarity among the users in the social network;
the trust degree unit is used for calculating the trust degree between the users according to the scoring matrix and the social network;
the first calculation unit is used for calculating the score similarity based on the user and the project similarity based on the behavior according to the score matrix;
the second calculation unit is used for calculating the semantic similarity of the project according to the knowledge graph;
the third calculation unit is used for calculating the item fusion similarity according to the item semantic similarity and the item similarity based on the behaviors;
the first fusion unit is used for calculating user fusion similarity according to the scoring similarity based on the users, the trust between the users and the social similarity between the users;
the second fusion unit is used for fusing the project fusion similarity and the user fusion similarity into a basic matrix decomposition model to obtain a recommendation model fusing the social network and the knowledge graph;
and the prediction unit is used for minimizing the recommendation model fusing the social network and the knowledge graph according to a gradient descent method to obtain the prediction score of the user on the unscored target item.
10. A recommendation device fusing social networks and knowledge graphs, the device comprising a processor and a memory:
the memory is used for storing program codes and transmitting the program codes to the processor;
the processor is configured to execute the method for fusing social networks and knowledge-graphs according to any one of claims 1-8 according to instructions in the program code.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110099046.3A CN112765488A (en) | 2021-01-25 | 2021-01-25 | Recommendation method, system and equipment fusing social network and knowledge graph |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110099046.3A CN112765488A (en) | 2021-01-25 | 2021-01-25 | Recommendation method, system and equipment fusing social network and knowledge graph |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112765488A true CN112765488A (en) | 2021-05-07 |
Family
ID=75707268
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110099046.3A Pending CN112765488A (en) | 2021-01-25 | 2021-01-25 | Recommendation method, system and equipment fusing social network and knowledge graph |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112765488A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113360784A (en) * | 2021-06-22 | 2021-09-07 | 北京邮电大学 | Collaborative filtering algorithm for knowledge graph optimization recommended by equipment operation and maintenance scheme |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109408734A (en) * | 2018-09-28 | 2019-03-01 | 嘉兴学院 | A kind of collaborative filtering recommending method of fuse information Entropy conformability degree and dynamic trust |
CN111125540A (en) * | 2019-11-12 | 2020-05-08 | 华侨大学 | Recommendation method integrating knowledge graph representation learning and bias matrix decomposition |
-
2021
- 2021-01-25 CN CN202110099046.3A patent/CN112765488A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109408734A (en) * | 2018-09-28 | 2019-03-01 | 嘉兴学院 | A kind of collaborative filtering recommending method of fuse information Entropy conformability degree and dynamic trust |
CN111125540A (en) * | 2019-11-12 | 2020-05-08 | 华侨大学 | Recommendation method integrating knowledge graph representation learning and bias matrix decomposition |
Non-Patent Citations (1)
Title |
---|
邓志彬等: "融合社交网络用户相似度的社会化推荐", 《计算机科学与应用》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113360784A (en) * | 2021-06-22 | 2021-09-07 | 北京邮电大学 | Collaborative filtering algorithm for knowledge graph optimization recommended by equipment operation and maintenance scheme |
CN113360784B (en) * | 2021-06-22 | 2023-09-19 | 北京邮电大学 | Collaborative filtering algorithm for knowledge graph optimization of equipment operation and maintenance scheme recommendation |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Hussain et al. | q‐Rung orthopair fuzzy soft average aggregation operators and their application in multicriteria decision‐making | |
CN111428147B (en) | Social recommendation method of heterogeneous graph volume network combining social and interest information | |
Zhang et al. | User community discovery from multi-relational networks | |
CN109389151B (en) | Knowledge graph processing method and device based on semi-supervised embedded representation model | |
CN109190030B (en) | Implicit feedback recommendation method fusing node2vec and deep neural network | |
CN112507246B (en) | Social recommendation method fusing global and local social interest influence | |
Peng et al. | N-dimensional Markov random field prior for cold-start recommendation | |
CN113744032B (en) | Book recommendation method, related device, equipment and storage medium | |
Nobahari et al. | ISoTrustSeq: a social recommender system based on implicit interest, trust and sequential behaviors of users using matrix factorization | |
CN112131261B (en) | Community query method and device based on community network and computer equipment | |
CN113918834B (en) | Graph convolution collaborative filtering recommendation method fusing social relations | |
CN113918832A (en) | Graph convolution collaborative filtering recommendation system based on social relationship | |
CN110781405B (en) | Document context perception recommendation method and system based on joint convolution matrix decomposition | |
Neve et al. | Hybrid reciprocal recommender systems: Integrating item-to-user principles in reciprocal recommendation | |
Albanese et al. | Modeling recommendation as a social choice problem | |
Amara et al. | Cross-network representation learning for anchor users on multiplex heterogeneous social network | |
CN104035978B (en) | Combo discovering method and system | |
CN112765488A (en) | Recommendation method, system and equipment fusing social network and knowledge graph | |
Bi et al. | Hierarchical social recommendation model based on a graph neural network | |
CN110795640B (en) | Self-adaptive group recommendation method for compensating group member difference | |
Stankova et al. | Classification over bipartite graphs through projection | |
CN112257959A (en) | User risk prediction method and device, electronic equipment and storage medium | |
Zhao et al. | A Hierarchical Attention Recommender System Based on Cross‐Domain Social Networks | |
Shilin | User model-based personalized recommendation algorithm for news media education resources | |
CN109658277A (en) | A kind of science social networks friend recommendation method, system 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 |