Hybrid Recommendation Method Based On Memory Network And Sentiment Analysis

Recommender system,as an important part of e-commerce platform,can effectively capture the user's interest characteristics and achieve personalized recommendation tasks.As an important branch of recommendation system,sequential recommendation system can further capture the dynamic changes of user interest characteristics and realize real-time recommendation task.Although the classical sequential recommendation methods can provide better recommendation services for users,this paper argues that these methods still have some defects.On the one hand,these methods are often difficult to capture the complex transformation relationship between items in the user behavior sequence and users' different degrees of concern for different items when modeling user interest.On the other hand,these methods often ignore the interference effect of user negative emotion on interest modeling.To solve these problems,a hybrid sequential recommendation model based on memory network and sentiment analysis is proposed in this paper.1)In order to capture the complex transition relationship between items in the user behavior sequence and users' different degrees of concern for different items,this paper uses the memory network based on attention mechanism as the recommendation module.Firstly,the method constructs the user behavior sequence into the corresponding session graph.Secondly,the constructed session graph is used as the input data of the memory network to generate the corresponding vector representations of different nodes.Then,the attention mechanism is used to capture the intrinsic different degrees of concern of users' interest and the external influence of target items on the modeling of users' interest,so as to obtain the stable long-term interest preference of users.Finally,the dynamic interest model of users is constructed based on the long-term preference of users and the short-term preference of the last item in the user behavior sequence.2)In order to effectively reduce the interference of user negative emotions on user interest modeling,this paper adopts sentiment analysis pre-filtering module to conduct sentiment analysis on comment text,so as to improve the data validity of user behavior sequence.Firstly,the method constructs a textual vector representation of user comments through Word2 Vec and TF-IDF.Then,LSTM is used to conduct sentiment analysis on user comment text,and the sentiment index of user comment text is obtained.Finally,the results of comment text sentiment analysis are used to filter the user behavior sequence,so as to improve the data validity of the memory network recommendation module based on attention mechanism,and finally improve the overall recommendation effect of the sequential recommendation model.By combining the emotion pre-filtering module of user comments with the memory network recommendation module based on attention mechanism,this method can improve the data validity of user behavior sequence,comprehensively capture the user's interest and preference characteristics from many aspects,and improve the recommendation effect of existing sequential recommendation methods to a certain extent.