Research On Recommendation Algorithm Based On Graph Neural Network Combined With Comment Text Analysis

With the development of network technology,the phenomenon of information explosion appears,which makes users look for a needle in a haystack when obtaining the information they need.Information explosion not only makes it difficult for users to obtain information,but also leads to the extreme sparsity of recommendation data.Personalized recommendation,as a method that can help users filter massive information and quickly obtain needed content,came into being.However,compared with a large number of users and items,there is little user item interaction information,which greatly affects the accuracy of recommendation.The traditional recommendation algorithm is not enough to meet the personalized needs of people at present.In the recommendation system,most of the data essentially has a graph structure.With the development of deep learning technology and graph neural network technology,graph structure data can be processed effectively,and personalized recommendation has ushered in a new research direction.However,because graph neural network directly deals with the abstract expression of relationship,the recommended results sometimes do not accord with human understanding,so the interpretability is poor.In this paper,a graph neural network recommendation algorithm model(GREA)based on random walk embedding combined with sentiment analysis is proposed,and in order to further improve the recommendation effect,GREA is improved,and the optimized model GREAs is obtained.Among them,GREA has done the following model construction and research based on the specific analysis of the problems raised above:(1)In view of the problems of poor scalability and poor feature extraction results caused by the complex feature extraction methods in traditional recommendation algorithms,graph neural network technology is used for feature extraction.In this paper,a graph structure can be constructed according to the user-item interaction mode,the interaction information is modeled,and the features of the nodes in the interaction graph are extracted by using the graph neural network technology.(2)Aiming at the problem of poor interpretability in the recommendation system based on graph neural network,it is proposed to combine the features in the review text information as auxiliary information to increase the interpretability of the recommendation results.This paper considers that the part of the comment text that affects the user is the semantic and emotional information contained in the comment,so by analyzing the sentiment polarity of the comment content and introducing the attention mechanism,we can obtain the relationship between each node and the emotional and semantic information.Personalized emotional traits.(3)Aiming at the problem of data sparsity in recommender systems,two methods are proposed to alleviate data sparsity by introducing review text analysis and initial graph embedding based on random walk-based matrix factorization to generate graph neural networks.The improvement of the GREAs model based on the GREA model lies in the addition of a comment text content extraction module,and the introduction of an attention mechanism to improve the graph neural network and text feature extraction.When improving the sentiment analysis module,considering that only the shallow network was used as the downstream sentiment analysis network layer for the sentiment analysis task in the previous chapter,a deep neural network Bi-LSTM was proposed to further extract the contextual relevance of the embedded text.The accuracy of sentiment analysis further improves the accuracy of recommendation algorithms.Finally,the model GREA proposed in this paper and its improved model GREAs are verified on three real data sets,and they have achieved better results than the baseline model.