Research And Application Of Collaborative Filtering Recommendation Algorithm Based On GCN And LSTM

At a time when 5G technology is in a pervasive stage and streaming and shopping platforms are mushrooming,there is an urgent need for a tool to help users filter and filter to the point where they are being pushed the content they want to see.Recommendation algorithms shift the dominance of information interaction from user to platform,to platform to user.Collaborative filtering is a common and easy-to-implement recommendation algorithm,but traditional collaborative filtering algorithms suffer from data sparsity and the inability to make dynamic recommendations,resulting in low recommendation performance.In recent years,scholars have improved the traditional collaborative filtering algorithm by deep learning and achieved good results,so that deep learning plays an important role in solving the problems of sparse data and dynamic recommendations.Therefore,this thesis focuses on deep learning and traditional collaborative filtering algorithms to study recommendation algorithms.In the existing research in the past,K-Means algorithm is often used to solve the collaborative filtering data sparsity problem,but the effect is not obvious.In contrast,graph convolutional neural network(GCN)can extract many useful target node information for bipartite graphs,thus achieving data sparsity mitigation.Based on this,the long short-term memory network(LSTM)is used to extract the features of users’ short-term preferences to solve the problem that traditional collaborative filtering algorithms cannot make dynamic recommendations.The details of the research are as follows:(1)Research on collaborative filtering recommendation algorithm based on K-Means and GCN.To address the data sparsity problem that traditional collaborative filtering algorithms ignore the potential connection between users and items,a collaborative filtering recommendation algorithm based on K-Means and GCN(KGCF)is proposed.The algorithm first analyzes the user group,clusters the users by K-Means algorithm to get different user clusters,fills the "user-item" matrix and transforms it into graph structure data,and then uses GCN to summarize the neighboring messages from different order neighborhoods to extract the higher-order potential relationships of the nodes and embed them in the graph structure data.The higher-order potential relationships are extracted and embedded in the node relationships,and then aggregated in a hierarchical manner for collaborative filtering.Finally,a comparison experiment is conducted on the Movie Lens dataset to demonstrate the effectiveness of the algorithm.(2)Research on collaborative filtering recommendation algorithm based on GCN and LSTM.To address the problem that collaborative filtering algorithms cannot achieve dynamic recommendation,a collaborative filtering recommendation algorithm(GLCF)based on GCN and LSTM is proposed.Therefore,this algorithm is a complement to the KGCF algorithm for dynamic recommendation.This thesis integrates LSTM and attention mechanism for automatic weight assignment and uses Time-LSTM to extract the potential relationship between time interval and user interest,so as to achieve the effect of dynamic recommendation.Finally,a comparison experiment is conducted on two publicly available datasets,Movie Lens and Amazon-Book,to demonstrate that this algorithm outperforms other traditional algorithms in recommendation performance.(3)Design and implementation of a movie recommendation system based on GCN and LSTM collaborative filtering recommendation algorithm.The system includes backend management module,user module and recommendation algorithm module.The backend management module and the user module are developed based on Java EE;the GLCF algorithm is used as the recommendation algorithm,and the recommendation algorithm module interacts with the front and backend of the system via http based on Django framework.