CN115033803A - Social media user recommendation method based on meta-path - Google Patents
Social media user recommendation method based on meta-path Download PDFInfo
- Publication number
- CN115033803A CN115033803A CN202210538041.0A CN202210538041A CN115033803A CN 115033803 A CN115033803 A CN 115033803A CN 202210538041 A CN202210538041 A CN 202210538041A CN 115033803 A CN115033803 A CN 115033803A
- Authority
- CN
- China
- Prior art keywords
- node
- meta
- nodes
- path
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000005295 random walk Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 8
- 238000009499 grossing Methods 0.000 claims description 4
- 238000010801 machine learning Methods 0.000 claims description 3
- 238000007619 statistical method Methods 0.000 abstract description 2
- 238000004364 calculation method Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 239000000126 substance Substances 0.000 description 3
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000013106 supervised machine learning method 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
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/958—Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
-
- 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)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a meta-path-based social media user recommendation method, which abstracts a social network into a heterogeneous information network, measures the association strength between each node in the network and recommended content through a statistical method, then limits random walk of the meta-path by taking the association strength as a weighted value of the node, and finally fuses the recommendation results of the single nodes to obtain a final recommendation result. The invention considers more information in the network in the recommendation process, and can obtain more accurate correlation metric value, thereby obtaining more accurate recommendation result.
Description
Technical Field
The invention belongs to the technical field of internet, and particularly relates to a user recommendation method.
Background
With the popularization of the internet and the arrival of the big data era, social media platforms such as Facebook, twitter and microblog have gradually become a part of people's lives. Thanks to the rise of the social media, information can be spread faster in the social media compared with the traditional media, and a user as a receiving party can also receive more extensive information, so that the social media becomes a main source for people to obtain information. With the rapid development of social media, the social media are full of noise, and a large amount of false redundant information is spread in the social media, so that a user is very difficult to acquire required information from the social media facing massive information. Therefore, making targeted recommendations for users in social media is one of the important directions for social network research.
The user recommendation methods in the existing social network can be roughly divided into two types: a method for network-token-based learning and a method for meta-path-based learning.
The method based on the network representation learning is characterized in that the concept of the information network is introduced into the similarity search problem, the social network is abstracted into the information network, a node sequence is generated by random walk, therefore, the representation learning is carried out on the nodes in the information network to obtain a representation vector, and the recommendation result can be obtained by analyzing the representation vector. However, this kind of method only considers the local structure of the information network and does not consider the integrity of the network, so that the information on the network cannot be completely retained.
The meta-path-based method is to abstract a social network into a heterogeneous information network, the meta-path can be regarded as a path type of the network, different meta-paths contain different semantic information and represent different connections among nodes, and the relevance between the nodes in the semantic environment represented by the different meta-paths is measured by extracting the meta-paths in the network, random walk, representation learning and other methods, so that the relevance under the whole network is obtained, rich semantic information in the network is fully utilized, and the user recommendation problem is solved. However, in the process of analyzing the network, the nodes of the same type in the heterogeneous information network are not distinguished, so that the information in the network can not be completely reserved; and due to the complexity of a heterogeneous information network abstracted by a social network in actual operation, when the method is used, the nodes of the same type are not distinguished, so that the recommendation result obtained by the method usually loses part of information, the quality of the recommendation result is not high, and an inaccurate recommendation result is often obtained.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a social media user recommendation method based on a meta path.
The specific technical scheme of the invention is as follows: a social media user recommendation method based on meta-paths comprises the following steps:
s1, extracting information in a social network, abstracting the social network into a heterogeneous information network G ═ V, E >, wherein V represents a node set, E represents an edge set, and each heterogeneous information network corresponds to a node mapping function phi, V → A and an edge mapping function psi, E → R, wherein A is a node type set, R is an edge type set, and | A | + | R | > 2; in the constructed heterogeneous information network, R is a binary relationship, denoted as R (e, e ') if node e is connected to node e' by a relationship R, and R (e) { e ': R (e, e') }, denotes a set of nodes connected to node e by a relationship R, for one meta-pathDomain of definition of meta-pathway dom (p) ═ dom (R) 1 )=A 1 Range (R) of the meta path l )=A l+1 ;
S2, randomly marking a part of nodes in the heterogeneous information network formed in the step S1, keeping the number of marked nodes related to and unrelated to the recommended content consistent during marking, and then measuring the association strength between each node in the network and the recommended content;
s3, taking the association strength obtained in the step S2 as a weight value of each node, taking the node related to the recommended content as a starting node to limit random walk of the meta-path, and measuring the association degree between each node and the starting node about one meta-path;
s4: constructing a training set according to the result obtained in the step S3, obtaining a weight value of each meta-path in the heterogeneous information network by using supervised machine learning, and performing weighted summation on the result obtained in the step S3 to obtain the association degree of each node and the initial node;
s5: and (4) fusing the single-node recommendation results obtained in the step (S4) to obtain a multi-node recommendation result by measuring the influence degree of each starting node in the heterogeneous information network.
Further, the association strength in step S2 is specifically represented by the CS score of the node, and specifically takes the lower bound of the Wilson score interval as the CS score of the user:
wherein CS (u) i ) Representing node u i The CS score of (a) is,representing node u i Marking the node proportion, n, related to the topic in the neighbor nodes i Representing a node u i Is marked with the number of neighbor nodes, z α Then the Z score representing the confidence level of the corresponding alpha;
wherein c represents a node u i Number of labeled neighbor nodes, n, associated with the query content i Representing node u i M represents the smoothing coefficient.
The invention has the beneficial effects that: according to the method, the social network is abstracted into a heterogeneous information network, the association strength between each node in the network and the recommended content is measured through a statistical method, then the association strength is used as the weight value of the node to carry out random walk of the limiting element path, and finally the recommendation results of the single nodes are fused to obtain the final recommendation result. The invention considers more information in the network in the recommendation process, and can obtain more accurate correlation metric value, thereby obtaining more accurate recommendation result.
Drawings
Fig. 1 is a schematic flow chart of a user recommendation method in an embodiment of the present invention.
Fig. 2 is a schematic diagram of a network mode in an embodiment of the present invention.
Detailed Description
The embodiments of the present invention will be further described with reference to the accompanying drawings.
In the heterogeneous information network, the meta-path contains different semantic information as the path type, if two nodes are connected through a certain meta-path, the two nodes have certain relevance with respect to the semantic information, and the relevance between the nodes in the heterogeneous information network can be obtained by measuring the relevance between the two nodes under the semantic information represented by the different meta-paths. The essence of recommendation is to recommend the node with the maximum relevance with the recommended content, so the method used by the invention can effectively solve the user recommendation problem in social media. The specific process of the invention is shown in fig. 1, and comprises the following steps:
s1, extracting information in a social network, and abstracting the social network into a heterogeneous information network G (V, E), wherein V represents a node set, E represents an edge set, and each heterogeneous information network corresponds to a node mapping function phi V → A and an edge mapping function psi E → R, wherein A is a node type set, R is an edge type set, and | A | + | R | > 2; in the constructed heterogeneous information network, R is a binary relationship, denoted as R (e, e ') if node e is connected to node e' by a relationship R, and R (e) { e ': R (e, e') }, denotes a set of nodes connected to node e by a relationship R, for one meta-pathDomain of definition of meta-pathway dom (p) ═ dom (R) 1 )=A 1 Range (R) of the meta path l )=A l+1 ;
S2, marking a part of nodes (related to and unrelated to recommended content) in the heterogeneous information network formed in the step S1, and when marking, keeping the number of marked nodes related to and unrelated to the recommended content consistent, and measuring the association strength between each node and the recommended content in the network;
s3, taking the association strength obtained in the step S2 as a weight value of each node, taking the node related to the recommended content as a starting node to limit random walk of the meta-path, and measuring the association degree between each node and the starting node about one meta-path;
s4: constructing a training set according to the result obtained in the step S3, obtaining a weight value of each meta-path in the heterogeneous information network by using supervised machine learning, carrying out weighted summation on the result obtained in the step S3 to obtain the association degree of each node and the initial node, wherein the nodes with higher association degrees are more in line with the recommendation requirement;
s5: by measuring the influence degree of each initial node in the heterogeneous information network, the single-node recommendation results obtained in the step S4 are fused to obtain multi-node recommendation results, so that the recommendation results are more accurate and reasonable.
In the social network, users pay attention to the users who are interested in the users, and the content published by the users has large relevance with the attributes of the users, so people and fans which are interested in the users and text content published by the users are very interested in building a heterogeneous information network. Aiming at text contents published by a user in social media, the embodiment extracts four types of information of Url, Hashtag, @ and Ner, wherein Url is a link of videos, webpages and the like which are published by the user and interested by the user, Hashtag is a topic involved in discussion by the user, @ is a user mentioned when the user publishes the text contents, and Ner is a named entity such as a name of a person, a place name and a professional noun which are identified for the text contents published by the user by using a tool provided by Stanford. Finally, a heterogeneous information network based on a social network is constructed, wherein the network comprises five types of nodes including User, Url, Hashtag, Mention and Ner, and edges including two types of Follow and Post.
In the process of processing heterogeneous information networks, due to the complexity of the networks and the overlarge data volume, the heterogeneous information networks are difficult to processThe network is described and processed, and therefore the heterogeneous information network is described using its meta-structure, i.e., its network model. The network mode is a minimum representation form of the heterogeneous information network, is an element structure of the heterogeneous information network G ═ V and E ≧ V, and comprises a node mapping phi V → A and an edge mapping phi E → R, the node is a defined node type A, and the relationship connecting the nodes is a defined edge type R, represented as T G (a, R). The network mode shown in fig. 2 is the network mode of the established heterogeneous information network. By constructing the network model, it can be found that nodes in a heterogeneous information network can be the same through different paths, for example: the User and the User can be connected through two paths of the User-User and the User-Url-User, so that the semantic information represented by the two paths is quite different, the paths with different semantic information are meta paths, and in a heterogeneous information network, the meta paths with different semantic information in a table can be replaced according to requirements so as to take the semantic information into a recommendation result.
The method discusses searching for a user that best matches a group of users who are interested in a topic based on the group. In most previous methods using heterogeneous information networks, nodes of the same node type are not distinguished. In this embodiment, in order to distinguish nodes of the same node type, the CS scores of the nodes in the network are calculated, where the CS scores represent the association strength between the nodes and topics, and the CS scores of the nodes are measured, so that the nodes of the same type can be distinguished in the similarity measurement.
The CS score of a node is calculated by a direct method, which is to mark all neighboring nodes and then obtain the proportion of nodes related to a topic in the neighboring nodes. However, because the number of the social network nodes is large, the information is huge, and it is obviously unreasonable to manually mark all the nodes. In order to solve the problem, a part of nodes in the heterogeneous information network are marked, and the CS fraction of the marked nodes in the neighbor nodes relative to the topic is used as the marked nodes in the neighbor nodes, so that the association strength between the user and the topic can be reflected to some extent.
However, the above method cannot distinguish the number difference of the neighboring nodes marked by different nodes, for example: two nodes u 1 And u 2 Wherein u is 1 There are 100 labeled neighbor nodes and related to topic, u 2 There are 1 labeled neighbor nodes associated with the query content, if the CS scores of both nodes are the same according to the above method, but the result is obviously unreasonable. To distinguish this case, the present embodiment introduces a confidence interval based on Wilson score to measure the CS score of the node, where the lower bound of the Wilson score interval is taken as the CS score of the user, as shown in equation (1):
wherein the content of the first and second substances,representing node u i Marking the node proportion, n, related to the topic in the neighbor nodes i Representing node u i Marks the number of neighbor nodes, and z α Then the Z-score corresponding to the alpha confidence is indicated. Based on equation (1), u can be calculated 1 And u 2 The CS scores of (1) are 0.97 and 0.23 respectively, and it can be seen that after the Wilson score confidence interval is introduced, u can be effectively distinguished 1 And u 2 The difference between them.
Another problem is found after using the above method, assuming two nodes u 1 And u 2 Wherein u is 1 There are 100 labeled neighbor nodes and all are topic independent, u 2 There are 1 marked neighbor users that are irrelevant to the topic, and both cases result in that the lower bound of the Wilson score is 0, which is obviously not reasonable enough. To distinguish this case, in the calculationIntroduction of Laplace's heimSlip, as shown in equation (2):
wherein c represents a node u i Number of labeled neighbor nodes, n, associated with the query content i Representing node u i M represents a smoothing coefficient, and m is set to 0.5 to obtain u 1 And u 2 The corresponding CS scores were 0.0005 and 0.027, respectively.
By introducing Wilson scores and Laplace smoothing, the finally calculated CS scores can intuitively reflect the association strength between the nodes and the topics, the obtained CS scores are distributed between 0 and 1, the higher the CS scores are, the stronger the association strength between the nodes and the topics is, and the weaker the association strength between the nodes and the topics is otherwise.
Recommending a group of user nodes in a heterogeneous information network and carrying out recommendation on any given meta-pathAnd a corresponding query node e search For simplicity of description, it is abbreviated as P ═ R 1 R 2 ...R l Then let P' ═ R 1 R 2 ...R l-1 At this time, the probability (degree of association between each node and the start node with respect to one meta path) to the node e in the heterogeneous information network is distributed as shown in equation (3):
wherein the content of the first and second substances,representing node e' and starting node e search Probability value of node e 'to node e' connected by meta-path P search With respect to the relevance of the meta-path P', CS (n) represents the CS score of the node n, and I (e) is an indicator function, in particularAs shown in formula (4):
wherein R is l (e, e') represents the relationship R between node e and node e l Are connected. After the meta-path is given, the final probability distribution can be obtained by layer-by-layer iteration of formula (3).
In the above, the probability distribution is calculated for a specific meta-path, and in addition, multiple meta-paths can be extracted from the heterogeneous information network so as to measure the relevance from the perspective of multiple semantic information, in this embodiment, only the meta-path with the length l < 3 is selected to ensure that the selected meta-path contains enough information. When in the computing node e and the query node e search When the degree of association between nodes is high, it is necessary to ensure the node e search Is the domain of definition of the meta-path, i.e. dom (p) ═ phi (e) search ) And the node type of the node needs to be the value range of the meta-path, i.e. range (p) ═ phi (e), and after adding the above-mentioned restriction, a meta-path set is obtainedThe meta-path in this set is for query node e search Meta-paths that need to be considered. Finally obtaining node e and inquiring node e search The degree of correlation between them is shown in the formula (5)
Wherein, theta P Representing the weight of the meta-path P in the heterogeneous information network.
The matrix form of the above formula is shown in formula (6)
s=Aθ (6)
Wherein s is a sparse column vector of the association degree between the nodes and the query node in the heterogeneous information network, and theta is a column vector of a weighted value of the meta-path in the networkA is called a feature matrix, and the ith row of A is represented as A i 。
For the weight of the meta path in the heterogeneous information network, the method used in this embodiment is L-BFGS and a binomial log-likelihood loss function, which are obtained by a supervised machine learning method.
The training set in step S4 may be expressed as D { (e) } (m) ,r (m) ) 1, wherein M denotes the number of training sets, e (m) Is a query node, r (m) Is a vector representing the relationship with the query node, if nodes e and e (m) With a relationship of r (m) Otherwise, it is 0. After determining the training set, we can get the optimal meta-path weight value by maximizing equation (7)
Where λ is the regularization parameter, o (m) (θ) is an objective function for a single training sample, θ is a column vector of weight values of the meta path in the network, and a binomial log likelihood loss function is used here, which is specifically shown in equation (8):
wherein for the training sample (e) (m) ,r (m) ),A (m) Is a matrix of the characteristics of the same,representing a set of entities associated with the querying node,representing a set of entities that are not related to the querying node,wherein the content of the first and second substances,is a matrix A (m) Ith row of (1), theta T Is a transposed vector of theta, the (x) th being 1/(1+ e) x )。
The gradient of formula (8) can be obtained by calculation as shown in formula (9):
through the calculation, the weight value of the extracted meta-path in the heterogeneous information network can be finally obtained.
Through the calculation, the association degree between the user and the single starting node in the heterogeneous information network can be obtained, and the search result can be more reasonable by fusing the results obtained by the single starting node. Therefore, the influence of the initial node in the heterogeneous information network is calculated, and the correlation result of the multiple initial nodes is obtained.
Taking the number of neighbor nodes of the node as a factor for measuring the influence of the node, as shown in formula (10):
wherein, the table of N (e) is the neighbor node of the node e. When a group of users interested in the same topic is given, i.e. all the start nodes E ═ { E } 1 ,e 2 ,...,e n And then, obtaining influence of all the starting nodes on the network, thereby obtaining the association degree between each node and all the starting nodes in the heterogeneous information network, as shown in formula (11):
wherein, c (e) * ) Indicating a start node e * The impact on the heterogeneous information network,representing node e and node e * The degree of association between them. And calculating the association degrees between all the nodes and the initial node in the heterogeneous information network so as to obtain a recommendation result.
The method comprises the steps of firstly extracting required information from a social network, abstracting the information into a heterogeneous information network, giving out how to measure the association strength between nodes and recommended contents in the network, then using the association strength as the weight of the nodes to carry out random walk of a limiting element path to obtain the recommendation result of a single node, and finally, in order to take more information in the heterogeneous information network into consideration, giving out a method for measuring the influence weight of an initial node, and fusing the recommendation results of the single node to obtain the final user recommendation result.
The invention provides a social media user recommendation method based on meta-paths, and particularly provides a plurality of methods and paths for realizing the technical scheme, wherein the methods and the paths are the preferred embodiments of the invention; the scope of the invention is not limited to the specific illustrations and examples. Those skilled in the art may also make modifications and variations that do not depart from the spirit of the invention and these should also be construed as being protected by the present invention. All the components not specified in the embodiment can be realized by the prior art.
Claims (2)
1. A social media user recommendation method based on meta-paths comprises the following steps:
s1, extracting information in a social network, abstracting the social network into heterogeneous information networks G ═ V and E >, wherein V represents a node set, E represents an edge set, and each heterogeneous information network corresponds to a node mapping function phi, V → A and an edge mapping function psi, E → R, wherein A is a node type set, R is an edge type set, and | A | + | R | > 2; in the heterogeneous information network constructed, R is a binary relationship, denoted as R (e, e ') if node e is connected to node e' by a relationship R, and R (e) { e ': R (e, e') }, denotes a set of nodes connected to node e by a relationship RFor a meta path P:domain of definition of a predetermined meta-path, dom (p) ═ dom (R) 1 )=A 1 Range (R) of the meta path l )=A l+1 ;
S2, randomly marking a part of nodes in the heterogeneous information network formed in the step S1, wherein the quantity of marked nodes related to recommended contents and irrelevant nodes are required to be kept consistent during marking, and then measuring the association strength between each node in the network and the recommended contents;
s3, taking the association strength obtained in the step S2 as a weight value of each node, taking the node related to the recommended content as a starting node to limit random walk of the meta-path, and measuring the association degree between each node and the starting node about one meta-path;
s4: constructing a training set according to the result obtained in the step S3, obtaining a weight value of each meta-path in the heterogeneous information network by using supervised machine learning, and carrying out weighted summation on the result obtained in the step S3 to obtain the association degree of each node and the initial node;
s5: and (4) fusing the single-node recommendation results obtained in the step (S4) to obtain a multi-node recommendation result by measuring the influence degree of each starting node in the heterogeneous information network.
2. The method as claimed in claim 1, wherein the association strength of step S2 is specifically represented by CS scores of nodes, and the lower bound of Wilson score interval is specifically taken as the CS score of the user:
wherein CS (u) i ) Representing node u i The CS score of (a) is,representing a node u i Marking the node proportion, n, related to topics in neighbor nodes i Representing node u i Is marked with the number of neighbor nodes, z α Then the Z-score corresponding to the alpha confidence is expressed;
wherein c represents a node u i Number of labeled neighbor nodes, n, associated with the query content i Representing node u i M represents the smoothing coefficient.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210538041.0A CN115033803B (en) | 2022-05-18 | 2022-05-18 | Social media user recommendation method based on meta-path |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210538041.0A CN115033803B (en) | 2022-05-18 | 2022-05-18 | Social media user recommendation method based on meta-path |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115033803A true CN115033803A (en) | 2022-09-09 |
CN115033803B CN115033803B (en) | 2024-06-21 |
Family
ID=83121375
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210538041.0A Active CN115033803B (en) | 2022-05-18 | 2022-05-18 | Social media user recommendation method based on meta-path |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115033803B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115860880A (en) * | 2023-01-06 | 2023-03-28 | 中国海洋大学 | Personalized commodity recommendation method and system based on multilayer heterogeneous graph convolution model |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106709037A (en) * | 2016-12-29 | 2017-05-24 | 武汉大学 | Movie recommendation method based on heterogeneous information network |
CN112069415A (en) * | 2020-08-13 | 2020-12-11 | 中国海洋大学 | Interest point recommendation method based on heterogeneous attribute network characterization learning |
CN113094975A (en) * | 2021-03-22 | 2021-07-09 | 西安交通大学 | Method, system, equipment and storage medium for evaluating vulnerability of smart grid node |
CN113505294A (en) * | 2021-06-15 | 2021-10-15 | 黄萌 | Heterogeneous network representation recommendation algorithm fusing meta-paths |
CN114036826A (en) * | 2021-10-29 | 2022-02-11 | 深圳前海微众银行股份有限公司 | Model training method, root cause determination method, device, equipment and storage medium |
CN114510630A (en) * | 2021-12-31 | 2022-05-17 | 中南财经政法大学 | Academic paper recommendation method and system based on heterogeneous graph convolution neural network |
-
2022
- 2022-05-18 CN CN202210538041.0A patent/CN115033803B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106709037A (en) * | 2016-12-29 | 2017-05-24 | 武汉大学 | Movie recommendation method based on heterogeneous information network |
CN112069415A (en) * | 2020-08-13 | 2020-12-11 | 中国海洋大学 | Interest point recommendation method based on heterogeneous attribute network characterization learning |
CN113094975A (en) * | 2021-03-22 | 2021-07-09 | 西安交通大学 | Method, system, equipment and storage medium for evaluating vulnerability of smart grid node |
CN113505294A (en) * | 2021-06-15 | 2021-10-15 | 黄萌 | Heterogeneous network representation recommendation algorithm fusing meta-paths |
CN114036826A (en) * | 2021-10-29 | 2022-02-11 | 深圳前海微众银行股份有限公司 | Model training method, root cause determination method, device, equipment and storage medium |
CN114510630A (en) * | 2021-12-31 | 2022-05-17 | 中南财经政法大学 | Academic paper recommendation method and system based on heterogeneous graph convolution neural network |
Non-Patent Citations (2)
Title |
---|
XIA XIAO等: "A Novel Framework With Weighted Heterogeneous Educational Network Embedding for Personalized Freshmen Recommendation Under the Impact of COVID-19 Storm", 《IEEE ACCESS 》, no. 9, 26 April 2021 (2021-04-26), pages 1 - 11 * |
牛璐等: "基于不同搜索路径下成对随机游走的推荐算法", 《计算机集成制造系统 》, vol. 30, no. 4, 15 November 2021 (2021-11-15), pages 1 - 8 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115860880A (en) * | 2023-01-06 | 2023-03-28 | 中国海洋大学 | Personalized commodity recommendation method and system based on multilayer heterogeneous graph convolution model |
Also Published As
Publication number | Publication date |
---|---|
CN115033803B (en) | 2024-06-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Brochier et al. | Global vectors for node representations | |
Noekhah et al. | Opinion spam detection: Using multi-iterative graph-based model | |
TWI408560B (en) | A method, system and apparatus for recommending items or people of potential interest to users in a computer-based network | |
Zhang et al. | Mining users trust from e-commerce reviews based on sentiment similarity analysis | |
US10528574B2 (en) | Topical trust network | |
Rafiei et al. | A novel method for expert finding in online communities based on concept map and PageRank | |
Zhang et al. | STCS lexicon: Spectral-clustering-based topic-specific Chinese sentiment lexicon construction for social networks | |
CN110069713B (en) | Personalized recommendation method based on user context perception | |
KR100462542B1 (en) | contents search system for providing confidential contents through network and method thereof | |
Najafi et al. | Investigation of the correlation between trust and reputation in B2C e-commerce using Alexa ranking | |
KR101088710B1 (en) | Method and Apparatus for Online Community Post Searching Based on Interactions between Online Community User and Computer Readable Recording Medium Storing Program thereof | |
Xue et al. | Content-aware trust propagation toward online review spam detection | |
CN110321492A (en) | A kind of item recommendation method and system based on community information | |
O'Doherty et al. | Towards trust inference from bipartite social networks | |
Sun et al. | Conversational structure aware and context sensitive topic model for online discussions | |
Chen et al. | A topic-sensitive trust evaluation approach for users in online communities | |
CN115033803B (en) | Social media user recommendation method based on meta-path | |
Huang et al. | Optimizing Ontology Alignment through Improved NSGA‐II | |
Zeng et al. | Similarity from multi-dimensional scaling: Solving the accuracy and diversity dilemma in information filtering | |
CN114997723A (en) | Construction method of multi-source heterogeneous block chain quality evaluation model for enterprise operation activities | |
Sun | Topic modeling and spam detection for short text segments in web forums | |
Zhuo | Consumer demand behavior mining and product recommendation based on online product review mining and fuzzy sets | |
Kuang et al. | Community-based link prediction in social networks | |
Chaki et al. | Pattern mining approaches used in social media data | |
Niu et al. | Entity resolution with attribute and connection graph |
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 | ||
GR01 | Patent grant |