Research And Application Of Graph Learning Recommendation Algorithm Based On Multivariate Data Analysis

In today’s age of information overload,recommendation systems mine user preferences and habits from massive data to generate personalized recommendations for users,thereby alleviating the pressure of information overload.Current mainstream recommendation algorithms explicitly use preference data from user feedback,such as ratings and reviews,to predict user interests,but such data is often under-rich and static,limiting the quality of recommendations.The complexity of the multivariate data managed by recommendation systems makes the information it contains even more valuable and poses a significant challenge to the data analysis and recommendation work of recommendation systems.In addition,how to dynamically capture changes in user interests and reduce the inevitable noise interference in user behavior sequences are also challenges that need to be addressed in recommendation systems research.This thesis builds on existing research work to analyze and mine the information contained in users’ historical behavior to optimize recommendation algorithms.The main research work of this thesis is as follows:(1)In this thesis,a multi-view graph attention network is designed to model complex data and data relationships with the help of graph neural networks,adaptively process user attributes and behavioral data,and mine potential collaborative information between different item features.The method considers the importance of different item features from three perspectives: user longterm interest preferences,short-term interest preferences,and target query information,and finally integrates the information from the three perspectives to update each node’s representation of the graph through an attention mechanism to accurately encode user interest preferences.(2)To address the interference of noise in the user’s historical behavioral features to the user’s interest preference information,this thesis designs a graph pooling strategy to compress and extract the user interest features.Firstly,the global features are incorporated into the graph pooling method using a learnable content matrix to balance the gap between the user interest graph and the pooled graph.The importance of the source graph nodes in the clusters of the target pooled graph is then obtained by clustering-friendly distribution.Finally,the pooling of user interest feature graphs is achieved through a cluster assignment matrix.By working in concert with multi-view graph attention networks and graph pooling,this thesis presents the SR-MVG model,and conducts extensive experiments on five publicly available datasets to confirm the effectiveness of the model.(3)Based on the above research results,this thesis designs and implements an educational library resource recommendation system from the level of practical application.The system takes a university library borrowing data set as the data source and uses the SR-MVG model as the recommendation engine.In accordance with the requirement analysis,outline design,and detailed design,the code implements functions such as user login with separate authority,book resource recommendation,and user library borrowing data analysis,realizing the conversion between theoretical research and practical application.Finally,the functions of the system were also tested to ensure normal operation of the system.