Research On Recommendation Algorithm Of Review Sentiment Analysis Based On Deep Learning

In recent years,with the popularity of the Internet and the advancement of computer technology,the amount of information has grown exponentially.People enjoy the convenience brought by the Internet,but at the same time also suffer from fragmentation of information and information overload.The personalized recommendation system can not only meet the individual needs of users,improve the user experience,but also greatly improve the user's behavior conversion rate under specific business scenarios.It has become the mainstream method to solve the problem of information overload.At the same time,with the rapid development of artificial intelligence,deep learning is increasingly applied to the recommendation field.The interactive information generated and the reviews published by users during the purchase of products are all important data sources for the recommendation algorithms.The user's historical interaction behavior can clearly reflect the user's preference,and the reviews can express the user's specific motivation for purchasing the product.This paper starts to improve the performance of the review analysis recommendation algorithm from three aspects: analyzing the sentiment features of user reviews,modeling the features of user interaction,and feature fusion between interactive behaviors and review semantics.The research content is as follows:(1)A user sentiment analysis and recommendation algorithm based on a capsule network is proposed.By designing a shared review-aware attention mechanism,the semantic features of shared reviews can assist in extracting each user preferences and product attributes.In addition,it integrates all aspects of the user and the product and uses the capsule network to infer the user's overall sentiment towards the product from the fusion features.Experiments show that the algorithm can effectively improve the accuracy of rating prediction.(2)A multi-faceted auto-encoder-based interactive behavior modeling method is presented.By devising a multi-faceted auto-encoding framework to model the potential features of the user's multiple preference aspects,the user's various preferences can be captured more comprehensively.Furthermore,by designing an improved attention mechanism,multi-dimensional weights are used to replace traditional single weights to achieve a more fine-grained attention to each feature.In the experiment,the algorithm showed better performance in multiple evaluation metrics.(3)A hybrid recommendation algorithm based on reviews and interactive behaviors is proposed,and an interactive sequence-aware attention mechanism is designed.By using interactive sequence features to assist in extracting the aspects of user preference and product attribute,the effective integration of interaction sequence features and review sematic features are achieved.In addition,by adopting an improved dynamic routing mechanism to replace the traditional routing mechanism,the accuracy of the capsule network in inferring user sentiments is improved.Finally,based on the algorithm,a simulation product push system based on the MQTT protocol and the Android system environment is designed and implemented.