CN114154080A - Dynamic socialization recommendation method based on graph neural network - Google Patents
Dynamic socialization recommendation method based on graph neural network Download PDFInfo
- Publication number
- CN114154080A CN114154080A CN202111489428.3A CN202111489428A CN114154080A CN 114154080 A CN114154080 A CN 114154080A CN 202111489428 A CN202111489428 A CN 202111489428A CN 114154080 A CN114154080 A CN 114154080A
- Authority
- CN
- China
- Prior art keywords
- user
- project
- item
- social
- opinion
- 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 36
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 14
- 230000003993 interaction Effects 0.000 claims abstract description 36
- 230000006399 behavior Effects 0.000 claims abstract description 12
- 230000002776 aggregation Effects 0.000 claims description 27
- 238000004220 aggregation Methods 0.000 claims description 27
- 230000002452 interceptive effect Effects 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 230000007246 mechanism Effects 0.000 claims description 5
- 230000000694 effects Effects 0.000 abstract description 2
- 230000004931 aggregating effect Effects 0.000 abstract 1
- 230000007547 defect Effects 0.000 abstract 1
- 230000008901 benefit Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
- 230000000007 visual effect Effects 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
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)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a dynamic socialization recommendation method based on a graph neural network, and relates to the technical field of socialization recommendation systems. Although the existing recommendation method makes great progress, the existing recommendation method still has the defects that the historical behaviors of the user are not rich enough and all potential interests cannot be reflected; potential correlations among the extracted multiple items are ignored, resulting in a problem of information loss. The method comprises the steps of firstly, generating a dynamic user project graph based on historical behavior data and time stamps generated by user historical interaction; secondly, respectively aggregating the dynamic user project graph and the user social relationship to generate a user model; then, considering the potential similarity relation between the project nodes, and introducing the interaction and the opinion in the user project graph to generate a project model; and adjusting model parameters and performing preference prediction by integrating user and project model components. The method of the invention enables the items to better accord with the user preference so as to achieve the personalized recommendation effect and improve the recommendation accuracy.
Description
Technical Field
The invention relates to the technical field of social recommendation systems, in particular to a dynamic social recommendation method based on a graph neural network.
Background
With the development of the internet and the appearance of large-scale data, personalized recommendation enters the visual field of people as an indispensable information pushing method and is widely applied to various e-commerce platforms, so that the time and the energy for a user to select information of interest from massive data are reduced. Due to the development of social networks and the emergence of various social platforms, the method for recommending by utilizing the social relationship among users can improve the recommendation result by utilizing the relationships such as friends or trust, better simulate the recommendation process in the real society and better embody the role of people in the recommendation process, so the social recommendation becomes a research hotspot of industrial application.
Although great progress is made in the existing recommendation method, in reality, most users usually consume only few items and some users may be unwilling to evaluate the items, or due to benefit driving, the users may score the commodities unrealistically so that the data is unreal or cause sparse problems, so that the cold start problem is still caused by embedding the user item graph into the model only through the graph neural network; the historical behavior data of the user only comprises articles directly interacted by the user, is limited by system exposure and is often not rich enough, and cannot reflect all potential interests; potential dependencies between the extracted plurality of items are ignored, resulting in loss of information.
The main foundation of the social recommendation system is that users with certain connections have certain associations of interests. Therefore, the data in the social recommendation are divided into two different graphs, namely a social graph and a user item graph, and two different aggregation processing graphs are introduced, so that the purpose of learning the user representation from different angles is achieved. And fuse the user social connection with the two aggregations to solve the problem.
Disclosure of Invention
In view of the above, the present invention is directed to solve the problems in the prior art, and provides a dynamic social recommendation method based on a graph neural network, which can effectively find potential factors of users and items, enhance the recommendation accuracy, and solve the cold start problem.
The invention provides a dynamic socialization recommendation method based on a graph neural network, which comprises the following steps of:
s10: acquiring user project interaction data of each user, wherein the user project interaction data comprises project opinion scores, purchase time and social network relation of each user;
s20: dividing data into two different graphs, namely a dynamic user project graph and a social graph, learning user representation from different angles, and introducing two aggregations to obtain project space user potential factors and social space user potential factors to process the two graphs respectively;
s30: combining the two potential factors into a final user potential factor through a standard MLP (Multi-level Link protocol), acquiring the user potential factor, and modeling the user;
s40: according to an article difference correlation algorithm, calculating similarity relations between project nodes, introducing interaction and opinions in a user project graph, and modeling projects through standard MLP and potential factors of learning projects;
s50: the method comprises the steps of building a model by combining user and project modeling and optimizing an objective function to obtain a recommendation value of a user to a project, predicting the preference of the user according to the recommendation value generated by historical interaction data of the target user for any user, predicting the next project which is possibly interacted with the target user, and carrying out social recommendation.
Further, in step S20, the method includes the steps of:
s21: constructing a dynamic user item aggregation according to items interacted with by user history and opinions and time stamps of the items by the user, and obtaining a user latent factor h of the item space from the dynamic user item aggregationI i∈Rd；
S22: according to the relation of a user in a social platform, social aggregation among the users is performed on the users from the social perspective, however, the preference of the user is similar to social friends directly related to the user or is further influenced by the social friends strongly related to the user, a relationship attention mechanism is introduced to select representative social friends to describe the social information of the user, the potential factors of the user are further simulated by combining the social information, and the potential factor h of the social space user is obtained from the social aggregationS i∈Rd。
According to step S21, the interaction relation x of the user itemsiaThe formula is as follows:
wherein, during the interaction between the user and the project, the user can express his/her opinion in a scoring mode on the project, and the expression is r, r is formed by {1,2,3,4,5}, each user can score the project in a five-score mode, and the preference of the user on the project is captured through the opinion of the user on the project, so that an opinion embedding vector e is introducedr∈RdThen each view is represented by a dense vector; the user will also generate an interaction time in the process of interacting with the project, which is marked as taFor user uiAnd interaction with it with the opinion r and time taItem v ofaEmbedding items into q by multi-layered perceptron MLPaOpinion embedding erAnd time taBinding is carried out and is denoted gv。
Further, in step S40, the method includes the steps of:
s41: acquiring the times of purchasing each item, interactive behaviors and item opinion scores of each user;
s42: calculating similarity relation between project nodes according to a project difference correlation algorithm;
s43: combining the similarity relation of every two items between the same item and each item into similarity embedding based on the item;
s44: through standard MLP, combining project similarity with interaction and opinion embedding in a user project graph, learning latent factors of the projects, modeling the projects, and obtaining an interaction relation formula as follows:
wherein every two items have similarity relation, denoted as s, to form item similarity matrix embedding, denoted as esWill item vjEmbedding basic users into p through a multi-layer perceptron MLP with opinion r and item similarity stOpinion embedding erAnd embedding a similarity matrix esMake a knotIn combination, let us denote guIn combination with fjtRepresenting the interactive relationship of the user items in the item aggregate.
Further, in step S10, for a user and a commodity without a behavior, a graph neural network prediction model is first constructed through the user social attribute features and the commodity characteristics, and when the user has a purchasing behavior, the user is trained to model.
The invention has the beneficial effects that:
the dynamic project user graph aggregation method has the advantages that the key structural relationship is reserved by the aggregation of the dynamic project user graph through layer number compression, the representation of users or projects is enhanced, the problem of information loss caused by the fact that potential correlation among a plurality of interest projects extracted by the users is ignored is solved through project modeling, secondary or noisy structural data are removed, the graph structure can be conveniently and effectively learned, and therefore similar users can be better found for project recommendation.
The invention provides a principle method for jointly capturing interaction and opinions in the social graph of the user through modeling the social recommendation of rating prediction by the graph neural network model, and the opinion information plays an important role in improving the recommendation performance of the model.
The invention constructs a user modeling method, constructs a unified method to fuse multi-dimensional aggregation formed by various relationships such as user-project, user-user and the like, aggregates multi-neighbors of different types to contribute to node generation through an attention mechanism, and combines a plurality of graph neural networks to obtain potential factors of the user.
Drawings
FIG. 1 is a block diagram of a recommendation method of the present invention;
fig. 2 is a schematic diagram of the neural network structure.
Detailed Description
Referring to fig. 1-2, the present embodiment provides a dynamic social recommendation method based on a graph neural network, including the following steps:
s10: the user item interaction data of each user is acquired and comprises item opinion scores, purchase time and social network relations of each user.
S20: the data is divided into two different graphs, namely a dynamic user project graph and a social graph, user representation is learned from different angles, two aggregations are introduced, and the project space user latent factors and the social space user latent factors are obtained and are respectively processed.
The step S20 includes the steps of:
s21: and constructing a dynamic user project graph according to projects interacted with by the user history and opinions and time stamps of the projects by the user, and enhancing the dynamic representation of the user or the project by utilizing the collaborative information in the local subgraph related to the user or the project to form dynamic user project aggregation.
The purpose of dynamic user item aggregation is to know the potential factors of a user by considering three factors of items interacted with the user, time generated by the interaction and opinion of the user on the items, and the definition function is as follows:
wherein C (i) is related to user uiAssociated set of items, xiaTo represent a vector, user u is representediAnd item vaThere is an interaction between, AggreitemsIs a project aggregation function. σ is a nonlinear activation function, and W and b represent weights and biases of the neural network.
During a user's interaction with an item, the user can express his/her opinion in a scoring manner for the item, denoted as r, r ∈ {1,2,3,4,5}, each user can score the item on a pentagram basis, and capture the user's preference for the item through the user's opinion of the item.
Introducing opinion embedding vector er∈RdEach view is represented by a dense vector.
The user will also generate an interaction time in the process of interacting with the project, which is marked as ta. For user uiAnd interaction with it with the opinion r and time taItem v ofaEmbedding items into q by multi-layered perceptron MLPaOpinion embedding erAnd time taBinding is carried out and is denoted gvAnd using x in combinationiaThe interactive relation of the user items is represented, and the interactive relation formula is as follows:
s22: social aggregation between users is performed on users from a social perspective according to their relationships in a social platform, since the preferences of a user are similar to or influenced by social friends with which he/she is directly connected, and the strength of the relationship between users can further influence the behavior of the users from social relationships. Therefore, the user potential factor of social aggregation should consider the strong association advantage of social relations. An attention mechanism is introduced to select representative social friends, describe the users' social information, and aggregate their information, with the formula:
wherein N (i) is user uiI.e. with user uiPossessing a strongly associated social relationship, AggreneibhorsAn aggregation function for the user's neighbors.
S30: and combining the two potential factors into a final user potential factor through the standard MLP, acquiring the user potential factor, and modeling the user.
Obtaining project space user latent factor h from dynamic user project aggregationI i∈Rd。
Obtaining social space user latent factor h from social aggregationS i∈Rd。
Combining the two latent factors into a final user latent factor h through standard MLPiWherein the itemThe spatial user latent factors and the social spatial user latent factors are connected together before being input into the multi-layered perceptron MLP.
User latency factor hiThe definition formula is:
c2＝σ(W2·c1+b2)
...
hi＝σ(Wk·ck-1+bk)
where k is the number of hidden layers and W and b represent the weight and bias of the layers.
S40: and calculating similarity relation between project nodes according to an article difference correlation algorithm, introducing interaction and opinions in a user project graph, and modeling the project by learning latent factors of the project through standard MLP.
A method similar to learning item latent factors through dynamic user item aggregation is used, which is represented as follows:
wherein z isjIs a project latent factor. For each item vjFrom and vjThe aggregate information in the interactive user set is denoted as b (j). To the item v by the userjThe function of (b) point of view perception interaction aggregation to (b), (j) is denoted as Aggreusers。
The step S40 includes the steps of:
s41: and acquiring the times of purchasing each item, the interactive behavior and the item opinion score of each user.
Randomly ordering all items, and according to the formula:
pt＝|vi 1,...,vi j,...,vi m|T
establishing a user embedded p which corresponds to the jth user and has the number of items equal to the total number of itemstEach item of the initial vector is in one-to-one correspondence with each item. Where m represents the total number of items, j represents the j-th item after random sorting, j is an element (1,2, …, m), vi jRepresenting the interactive behavior of the ith user on the jth item, i.e. embedding p by the usertItem (ii) | · (& gtu)TIs a matrix transposition.
Taking the item value corresponding to the item purchased by the ith user in the initial vector as 1, taking the item value corresponding to the item not purchased by the ith user in the interactive behavior as 0, and obtaining the initial interactive behavior p of the user(i) t。
S42: and calculating similarity relation between the project nodes according to the project difference correlation algorithm.
The items are associated with the user project graph and the interactions and opinions in the user project graph should be captured together to understand the underlying factors of the items. According to the formula:
respectively obtain the average value of the purchased times of any item alpha
And the average of the number of times any item beta was purchased
Where n is the total number of users, kαTotal number of times, k, that item α was purchasedβIs the total number of times item beta is purchased.
According to the formula:
obtaining the similarity between any item alpha and any item betas (α, β), i.e., the pairwise item similarity relationship between any one item and each item. Where i is the ith user, pi,αNumber of times, p, item alpha is purchased for the ith useri,α>0。pi,βNumber of times, p, item β is purchased for the ith useri,β>0。
S43: combining the similarity relation of every two items between the same item and each item into similarity embedding based on the item;
s44: and (3) combining project similarity with interaction and opinion embedding in a user project graph through standard MLP, learning latent factors of the projects, and modeling the projects.
Different users may also express different opinions during user item interactions for the same item. These opinions from different users may help users capture features of the same item in different ways.
Every two items have similarity relation, expressed as s (alpha, beta), forming item similarity matrix embedding, denoted as esWill item vjEmbedding basic users into p through a multi-layer perceptron MLP with opinion r and item similarity stOpinion embedding erAnd embedding a similarity matrix esBinding is carried out and is denoted guIn combination with fjtThe interactive relation of the user items in the item aggregation is represented, and the interactive relation formula is as follows:
and introduces an attention mechanism, with fjtAnd q isjAs input, the importance weight of a user is distinguished through two layers of neural attention networks, and the strong social relation influence of dynamic user item interaction on item potential factors is captured.
S50: the method comprises the steps of building a model by combining user and project modeling and optimizing an objective function to obtain a recommendation value of a user to a project, predicting the preference of the user according to the recommendation value generated by historical interaction data of the target user for any user, predicting the next project which is possibly interacted with the target user, and carrying out social recommendation.
The latent factor h of the useriAnd the latent factor z of the projectjIn combination with the prediction of the user's preference, the formula is as follows:
g2＝σ(W2·g1+b2)
...
gk-1＝σ(Wk·gk-1+bk)
r'ij＝wT·gk-1
where k is the number of hidden layers, W and b represent the weight and deviation, r ', of the layer'ijIs user uiFor item vjPreference prediction of (3).
Estimating model parameters, and optimizing an objective function, wherein the formula is as follows:
where | O | is the number of predictions, rijIs user uiFor item vjTrue user item relationships.
In The article Fan W, Ma Y, Li Q, et al, graph neural networks for social recommendation [ C ]// The World Wide Web conference.2019: 417-:
TABLE 1 data of the results
The first column in table 1 is a data set, 40% of Amazon Review data set and JData data set are respectively selected, the second column is an evaluation index, and the third column is a common recommendation method, where MAE and RMSE represent recommendation prediction errors, and the smaller the errors, the better the recommendation effect is proved, and the closer the user is to the last item selection result.
Claims (4)
1. A dynamic socialization recommendation method based on a graph neural network is characterized by comprising the following steps:
s10: acquiring user project interaction data of each user, wherein the user project interaction data comprises project opinion scores, purchase time and social network relation of each user;
s20: dividing data into two different graphs, namely a dynamic user project graph and a social graph, learning user representation from different angles, and introducing two aggregations to obtain project space user potential factors and social space user potential factors to process the two graphs respectively;
s30: combining the two potential factors into a final user potential factor through a standard MLP (Multi-level Link protocol), acquiring the user potential factor, and modeling the user;
s40: according to an article difference correlation algorithm, calculating similarity relations between project nodes, introducing interaction and opinions in a user project graph, and modeling projects through standard MLP and potential factors of learning projects;
s50: the method comprises the steps of building a model by combining user and project modeling and optimizing an objective function to obtain a recommendation value of a user to a project, predicting the preference of the user according to the recommendation value generated by historical interaction data of the target user for any user, predicting the next project which is possibly interacted with the target user, and performing social recommendation;
2. in step S20, the method includes:
s21: constructing a dynamic user item aggregation according to items interacted with by user history and opinions and time stamps of the items by the user, and obtaining a user latent factor h of the item space from the dynamic user item aggregationI i∈Rd；
S22: according to the relation of a user in a social platform, social aggregation among the users is performed on the users from the social perspective, however, the preference of the user is similar to social friends directly related to the user or is further influenced by the social friends strongly related to the user, a relationship attention mechanism is introduced to select representative social friends to describe the social information of the user, the potential factors of the user are further simulated by combining the social information, and the potential factor h of the social space user is obtained from the social aggregationS i∈Rd；
3. In the step S21, the interaction relation x of the user item is characterizediaThe formula is as follows:
wherein, during the interaction between the user and the project, the user can express his/her opinion in a scoring mode on the project, and the expression is r, r is formed by {1,2,3,4,5}, each user can score the project in a five-score mode, and the preference of the user on the project is captured through the opinion of the user on the project, so that an opinion embedding vector e is introducedr∈RdThen each view is represented by a dense vector; the user will also generate an interaction time in the process of interacting with the project, which is marked as taFor user uiAnd interaction with it with the opinion r and time taItem v ofaEmbedding items into q by multi-layered perceptron MLPaOpinion embedding erAnd time taBinding is carried out and is denoted gv；
4. In step S40, the method includes:
s41: acquiring the times of purchasing each item, interactive behaviors and item opinion scores of each user;
s42: calculating similarity relation between project nodes according to a project difference correlation algorithm;
s43: combining the similarity relation of every two items between the same item and each item into similarity embedding based on the item;
s44: through standard MLP, combining project similarity with interaction and opinion embedding in a user project graph, learning latent factors of the projects, modeling the projects, and obtaining an interaction relation formula as follows:
wherein every two items have similarity relation, denoted as s, to form item similarity embedding, denoted as esWill item vjEmbedding basic users into p through a multi-layer perceptron MLP with opinion r and item similarity stOpinion embedding erAnd embedding a similarity matrix esBinding is carried out and is denoted guIn combination with fjtRepresenting the interactive relationship of the user items in the item aggregate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111489428.3A CN114154080B (en) | 2021-12-07 | 2021-12-07 | Dynamic socialization recommendation method based on graph neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111489428.3A CN114154080B (en) | 2021-12-07 | 2021-12-07 | Dynamic socialization recommendation method based on graph neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114154080A true CN114154080A (en) | 2022-03-08 |
| CN114154080B CN114154080B (en) | 2024-08-16 |
Family
ID=80453314
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111489428.3A Active CN114154080B (en) | 2021-12-07 | 2021-12-07 | Dynamic socialization recommendation method based on graph neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114154080B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114564652A (en) * | 2022-04-29 | 2022-05-31 | 江西财经大学 | Personalized gift recommendation method and system based on user intention and two-way preference |
| CN114936323A (en) * | 2022-06-07 | 2022-08-23 | 北京百度网讯科技有限公司 | Graph representation model training method and device and electronic equipment |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287335A (en) * | 2019-06-17 | 2019-09-27 | 桂林电子科技大学 | The personalized recommending scenery spot method and device of knowledge based map and user's shot and long term preference |
| CN112084427A (en) * | 2020-09-15 | 2020-12-15 | 辽宁工程技术大学 | Interest point recommendation method based on graph neural network |
| CN112380435A (en) * | 2020-11-16 | 2021-02-19 | 北京大学 | Literature recommendation method and recommendation system based on heterogeneous graph neural network |
| CN112507246A (en) * | 2020-12-13 | 2021-03-16 | 天津大学 | Social recommendation method fusing global and local social interest influence |
| CN112905900A (en) * | 2021-04-02 | 2021-06-04 | 辽宁工程技术大学 | Collaborative filtering recommendation algorithm based on graph convolution attention mechanism |
| CN113377656A (en) * | 2021-06-16 | 2021-09-10 | 南京大学 | Crowd-sourcing recommendation method based on graph neural network |
| US20210366024A1 (en) * | 2020-05-25 | 2021-11-25 | National University Of Defense Technology | Item recommendation method based on importance of item in session and system thereof |
-
2021
- 2021-12-07 CN CN202111489428.3A patent/CN114154080B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287335A (en) * | 2019-06-17 | 2019-09-27 | 桂林电子科技大学 | The personalized recommending scenery spot method and device of knowledge based map and user's shot and long term preference |
| US20210366024A1 (en) * | 2020-05-25 | 2021-11-25 | National University Of Defense Technology | Item recommendation method based on importance of item in session and system thereof |
| CN112084427A (en) * | 2020-09-15 | 2020-12-15 | 辽宁工程技术大学 | Interest point recommendation method based on graph neural network |
| CN112380435A (en) * | 2020-11-16 | 2021-02-19 | 北京大学 | Literature recommendation method and recommendation system based on heterogeneous graph neural network |
| CN112507246A (en) * | 2020-12-13 | 2021-03-16 | 天津大学 | Social recommendation method fusing global and local social interest influence |
| CN112905900A (en) * | 2021-04-02 | 2021-06-04 | 辽宁工程技术大学 | Collaborative filtering recommendation algorithm based on graph convolution attention mechanism |
| CN113377656A (en) * | 2021-06-16 | 2021-09-10 | 南京大学 | Crowd-sourcing recommendation method based on graph neural network |
Non-Patent Citations (2)
| Title |
|---|
| YIFAN ZHU等: "recommending learning objects through attentive heterogeneous graph convolution and operation-aware neural network", IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, vol. 35, no. 4, 8 November 2021 (2021-11-08), pages 4178 - 4189 * |
| 周欢等: "基于图神经网络和标签可重叠社区的社会化影视推荐模型", 情报理论与实践, vol. 44, no. 6, 23 December 2020 (2020-12-23), pages 164 - 170 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114564652A (en) * | 2022-04-29 | 2022-05-31 | 江西财经大学 | Personalized gift recommendation method and system based on user intention and two-way preference |
| CN114936323A (en) * | 2022-06-07 | 2022-08-23 | 北京百度网讯科技有限公司 | Graph representation model training method and device and electronic equipment |
| CN114936323B (en) * | 2022-06-07 | 2023-06-30 | 北京百度网讯科技有限公司 | Training method and device of graph representation model and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114154080B (en) | 2024-08-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11659050B2 (en) | Discovering signature of electronic social networks | |
| CN109902708B (en) | Recommendation model training method and related device | |
| CN111428147B (en) | Social recommendation method of heterogeneous graph volume network combining social and interest information | |
| Guan et al. | Matrix factorization with rating completion: An enhanced SVD model for collaborative filtering recommender systems | |
| Fang et al. | Multi-faceted trust and distrust prediction for recommender systems | |
| Davoudi et al. | Social trust model for rating prediction in recommender systems: Effects of similarity, centrality, and social ties | |
| CN109767301A (en) | Recommended method and system, computer installation, computer readable storage medium | |
| CN114817663B (en) | Service modeling and recommendation method based on class perception graph neural network | |
| Guo et al. | Cold start recommendation based on attribute-fused singular value decomposition | |
| CN114154080A (en) | Dynamic socialization recommendation method based on graph neural network | |
| Kim et al. | Building a web of trust without explicit trust ratings | |
| Li et al. | Trust agent-based behavior induction in social networks | |
| CN110598118A (en) | Resource object recommendation method and device and computer readable medium | |
| Roy et al. | Trustingness & trustworthiness: A pair of complementary trust measures in a social network | |
| CN116542742A (en) | Heterogeneous dynamic social recommendation method based on multiple relation types | |
| Shi et al. | Cross-domain variational autoencoder for recommender systems | |
| CN114139066A (en) | Collaborative filtering recommendation system based on graph neural network | |
| Serrano | Smart internet search with random neural networks | |
| Yao et al. | Ddin: Deep disentangled interest network for click-through rate prediction | |
| CN111429175A (en) | Method for predicting click conversion under sparse characteristic scene | |
| CN115660776A (en) | Content resource recommendation method, device and medium integrating depth matrix decomposition and multi-feature intersection | |
| Zhang | Research on collaborative filtering recommendation algorithm based on social network | |
| CN108830680A (en) | Personalized recommendation method, system and storage medium based on discrete disassembler | |
| Van Pham et al. | A novel approach of voterank-based knowledge graph for improvement of multi-attributes influence nodes on social networks | |
| CN114692012A (en) | Electronic government affair recommendation method based on Bert neural collaborative filtering |
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 | ||
| GR01 | Patent grant |