Research On Personalized Recommendation System Based On Text Sentiment Analysis

The development of computer science brings people a convenient and reliable life.As the world entered the Internet era of globalization,more and more Internet applications were created and used,changed people's existing way of life.The emergence of the recommendation system is closely related to the rise of Internet commerce and the development of personalized recommendation,which promotes and pushes the development of facial recognition and information delivery.The influence of the recommendation system is as significant as the multimedia revolution.In this paper,we will analyze and research the present situation of the Internet recommendation system and its involvement in various fields,as well as summarize the common key technologies and methods used in the collaborative recommendation system.In order to meet the recommendation system to the needs of the user behavior in deep excavating,the text puts forward a kind of sentiment analysis method based on semantics.First of all,based on semantic recognition and the feature of the n-gram model method we can improve the SVM classification model to distinguish between statements,questions and negatives in the context.Then retrieve SentiWordNet databases under different contexts to express parts of speech and their meanings to build an emotional dictionary.Finally,according to the keywords and contextual links to determine escape relationship and degree of emotion,the text by using the method of weighted average of the emotional tendency.Due to the large number of data of the recommendation system,the user similarity matrix is too complex,which requires a lot of calculation when the system updates the recommendation form.The community division is a good way to solve this problem,the use of information entropy to describe the data of user comment,according to the critical value of the average entropy and entropy increment for community division of users and items.Collaborative filtering of the target user according to the community similarity between users,to calculate the K most similar users.Based on the most interested item categories of K similar users of the target user,recommending the item of the categories to the target users.In this paper,the designed system adopts B/S model and Web Service technology,so that the systems can perform cross-platform interactive operation.System architecture is divided into five layers,including datasource layer,data management layer,database layer,analytical application layer and results display layer.Supporting of system layer guarantees the strong scalability and flexibility.Through test,this paper designed by recommendation system has accurate user analysis ability and good real-time operation ability,able to work steadily to complete the Internet of personalized recommendation.