Research On Dual Attention Deep Recommendation Model Combining Reviews And Its Helpfulness

With the rapid development of cloud computing,big data,Internet of things and other technologies,the types of Internet applications emerge in endlessly,which leads to the explosive growth of data scale and serious information overload.As one of the important means to solve "information overload",recommendation system has been widely used.But the traditional recommendation system has the problems of cold start and data sparsity.In addition to the rating information of the recommendation system,the review information contains rich user interest and project feature information,which is helpful to learn user and project representation more accurately.In recent years,the integration of reviews into recommendations has attracted more and more attention.However,how to automatically and effectively obtain deeper user and project review features,and how to better integrate reviews and rating and other multi-source heterogeneous data need further study.This paper has carried out research on review-based deep recommendation models and algorithms.The main tasks completed are as follows:(1)Existing review-based recommendation methods usually use the same model to learn the review representation of all users and items.However,different users have different preferences and different items have different characteristics.Therefore,for different user item pairs,the same word or similar reviews may have different informativeness and different importance.Based on this,this paper proposes a Dual Attention Recommendation Model with Reviews(DARM)that incorporates review information.The model identifies user information embedding and item review embedding to identify key information to build local attention,and uses item review features to construct user review features for key review recognition of interactive attention.Based on the review features extracted based on deep learning and dual attention mechanism,and the rating features obtained by modeling the rating information,the factor prediction machine technology was finally used to make the rating prediction.(2)Since different reviews have different informational qualities,and many reviews are noisy or even misleading,combining the helpfulness information of reviews can help make better use of reviews for recommendations.Based on this,based on the DARM model,this paper proposes a deep recommendation model DARM-Help with review helpfulness.This model calculates the attention weight of reviews by defining the helpfulness of the reviews and calculating their ratings in order to better extract the characteristics of the reviews of users and items.Since many reviews do not have a helpfulness,the model proposes a multi-task learning method that combines the helpfulness prediction of reviews and the prediction of ratings.(3)The experiments of the above models and algorithms were performed on real datasets of Amazon and Yelp to verify their recommended performance.The experimental results show that the DARM model and DARM-Help model proposed in this paper have lower MSE than other rating prediction algorithms based on review information,that is,have better recommendation performance,and the DARM-Help model has a better recommendation effect than the DARM model.