Research On Sentiment Analysis And Recommendation System Of User Reviews Based On RoBERTa

The development of Internet technology has led to exponential growth of network information,and the rapid screening of useful information from these vast amounts of information has also become a major problem.In recent years,with the development of deep learning technology and its application in various specific fields,recommendation systems,natural language processing and other technologies have also made considerable progress.Among them,the recommendation system is one of the main means of information filtering and filtering.In the face of complicated user information,the recommendation system provides better services for users according to their historical information,emotional preferences and other records,and accurately filters target items from massive data to achieve accurate recommendation.With the explosion of user data and item data,due to the lack of historical data,traditional recommendation often has problems such as decreased recommendation accuracy,cold start,sparse data,insufficient interpretability,and inaccurate grasp of user’s emotional preferences.In order to overcome these shortcomings and improve the performance of the recommendation system,researchers began to focus on the peripheral data left by users in the process of surfing the Internet,such as location information,social network information,conversation information Comment information and so on.These peripheral information has great value.It is very useful for the model to study the application of these information to the recommendation system in depth.Among them,comment information is the most accessible and intuitive information that reflects users’ emotional preferences.It is the user’s most direct feedback on the item.Learn the user’s emotion and item attributes from the comment text,and then process them through algorithms,which can solve a series of problems such as sparse data,cold start,and lack of interpretability.In this paper,we mainly do the following research from the perspective of comment text processing:(1)First of all,we study the background and significance of this topic,and analyze the development process of sentiment analysis and recommendation system of comment text,as well as the status quo of research at home and abroad;Secondly,it summarizes various classical model structures applied in this paper on text processing and deep learning.(2)Ping therefore conducted an experiment on the word embedding method of each comment text.Through the analysis and comparison of the experimental results,the best word embedding layer was selected to build the emotional analysis model and comment recommendation model of this article.(3)The comment text often contains rich emotional information.Digging into this information can well understand users’ emotional preferences,and the effect of recommendation is self-evident.However,the implicit emotional expression and semantic diversity in the text have caused some problems to the extraction of emotional features.Therefore,in this paper,we propose a new model of comment emotion analysis,RoBERta-BiLSTM-BiLSTM(R2BL).This model embeds words into the input comment text through the RoBERTa layer to generate dynamic and context-dependent word vectors,which can more effectively express the comment information;The two-layer BiLSTM network layer is used to calculate,capture the bidirectional semantics of the comment text,extract the implicit emotional feature information contained in the text semantics,and then use a sigmoid function to classify the emotional polarity in the comment to get the results.Finally,an experiment is carried out on IMDB,a film review data set,and the excellent performance of the model proposed in this paper is proved by comparing it with the emotional analysis models of various categories.(4)The traditional comment and recommendation model only uses the comment data to extract the feature information,and often ignores the user’s emotional expression in the comment text,which can well reflect the user’s preferences,so it will lead to the inaccurate recommendation results and the lack of interpretability of the model.In order to solve these problems,this paper proposes a deep recommendation system RIST,which integrates the emotional tendency of users’ comments.The RoBERTA model is used to represent the comment text,which is input into the two-layer BiLSTM network to extract the user’s implicit emotional expression and get the user’s emotional characteristics;At the same time,RoBERTA’s comment embedding vector is input into the convolutional neural network to extract feature information,and then the useful features of users and items are filtered through the collaborative attention mechanism network;Integrate the comment features and user emotion features,connect and interact through the MLP network,and finally make a rating prediction for the model.The model selects four sub-data sets of product reviews in the Amazon public data set and carries out comparative experiments with the baseline model.The experimental results show that the model in this paper can more accurately reflect the real preferences of users,and the recommendation performance is significantly improved compared with other baseline models.