Research On Personalized Recommendation Based On Sentiment Analysis

With the extensive application of e-commerce and the gradual maturity of Web2.0 technologies,more and more users are keen to share and publish reviews and opinions through the Internet.Any product or service will generate thousands of evaluation information.User Reviews become an increasingly important information carrier,and at the same time,the problem of information overload has become increasingly serious.The emergence of personalized recommendation system provides users with a tool to solve information overload,but the traditional recommendation technology only considers the user's overall rating of the item,but often ignores the user review information.User reviews often include subjective views of the user,which can reflect the user's preferences and sentimental preferences for each attribute of the project.Therefore,fully excavating and utilizing this information carrier will help solve the problems such as cold start,sparse data,and low recommendation accuracy in the recommendation process,thereby realizing more accurate personalized recommendation.In view of the above issues,this paper reviews the research status of sentiment analysis and personalized recommendation technology,and proposes a novel recommendation model based on sentiment analysis that combines the effects of user reviews on similarity.The similarity based on item ratings translates into the user's similarity based on the emotional score of the project feature level.This article's recommendation model is mainly divided into two parts:user comment sentiment analysis and project recommendation.The sentiment analysis section is divided into three processes:First,the method of extraction based on syntactic dependency is used to extract product feature words and corresponding emotional words from user reviews using four types of syntactic relations.Second,dictionary-based semantic similarity is used.Degree calculation method,which clusters feature words into K attribute features,calculates and calculates all user comment sets,obtains user's attention to different features,and builds a user preference model;thirdly,the intensity of emotional tendencies designed according to this article.The scoring method calculates the user's emotional score on project feature level.This process fully considers the co-occurrence word order problems of degree adverbs and negative words,and finally obtains the user-feature emotion scoring matrix.The project recommendation parts are mainly divided into two steps:first,calculating the similarity between users based on user-feature sentiment scores,and using K-means for user clustering to generate user groups,and obtaining "nearest neighbors" among groups based on similarity.Second,a method based on score prediction is used to generate a to-be-recommended set,which is integrated into a user preference model to optimize the recommendation order to generate a recommendation list.Finally,this paper uses "Jingdong Mall" website as a data application platform,and gives the application process of the model and the calculated value of each step in detail.The experimental results show that the model is of a good recommendation effect and applicability.