Font Size: a A A

Research On Deep Recommendation Algorithm Based On Social Network Graph Mining

Posted on:2022-03-06Degree:MasterType:Thesis
Country:ChinaCandidate:N WangFull Text:PDF
GTID:2518306332957899Subject:Software engineering
Abstract/Summary:PDF Full Text Request
Recently,data mining and predictive analysis technologies have been applied ubiquitously in our daily lives,and the rise of social networks has also brought us great convenience.As the user scale of social networks has increased exponentially,research related to it has gradually become popular recently.The behavior of users in social networks can have an impact on the behaviors of other users related to it,and the thoughts or behaviors of other users can be modified in a subtle way.Therefore,in addition to retaining users,it is also for the purpose of deep-seated information through the relationship between users.Digging and recommending relevant information to users is extremely important.Recommendations through social relationships can make it more convenient for companies to recommend relevant information and adjust operating strategies according to user needs.Therefore,recommendations made through social network impact analysis are researches with development potential and application value in recent years.With the rapid progress of Internet technology in recent years,graph data mining technology has also been developed by leaps and bounds.The graph neural network can naturally integrate node information and topological structure when processing image data,making it extremely powerful in graph data learning and information mining.Benefiting from the above-mentioned advantages of graph neural network,by transforming data into user-item graph and user-user graph,graph neural network data mining technology has great potential in social network recommendation applications.Social recommendation also faces many challenges.For example,social networks always involve multiple graphs,such as user-user social relationship graphs and useritem evaluation graphs;social relations have different advantages;how to deal with the interaction between users and items and their associations opinions etc.This paper takes the data set publicly available on the Internet as the research object.To solve the above problems,firstly use graph convolutional neural network,and integrate the method of project and user relationship aggregation,and jointly capture the interaction relationship and opinions in the user project graph to conduct experimental research.Model the user first,extract the feature matrix from the user-item graph,and extract the adjacency matrix as input based on the user-user social relationship graph,use graph convolutional neural network for learning;then model the item for learning the potential factors of the project,in order to consider interaction and viewpoints in the user project diagram,we introduce user aggregation,that is,to aggregate the user's viewpoints in project modeling;finally,through the joint integration of users and project modeling components,to predict the learning model parameters.Subsequently,improvements were made to the deficiencies of the graph convolutional neural network.According to the feature vector set learned from the social data set as input,the graph attention mechanism neural network was used to learn when modeling users;at the same time,user aggregation was also used to aggregate The user's point of view is used in project modeling;finally,the user hidden factor and the item hidden factor are connected to complete the recommendation task of scoring prediction.This paper conducts a comparative analysis of recommended algorithm models through experiments on open source social network data sets,using average absolute error(MAE)and root mean square error(RMSE)as evaluation criteria.The experimental results prove that the two algorithm models proposed in this paper a re compatible with each other.The traditional social recommendation algorithm model is improved to a certain extent.
Keywords/Search Tags:Social Network, Recommendation Algorithm, Graph Neural Network, Graph Convolutional Neural Network, Graph Attention Network
PDF Full Text Request
Related items