Research On Sentiment Analysis And Personalized Recommendation Of Product Reviews Based On Graph Database

Faced with a large number of Internet information, recommendation system can quickly help consumers to find the information which they are interested in, and help the information producers publish information to attract more users, it's a good solution to the problem of information overload. The traditional recommendation system analyzes the users' profile information, product ratings or historical behavior data in a simply way, to study the users' interest, and recommend products or information to the users to meet their needs. With the growing number of user information and user behavior data on the Internet, the traditional recommendation system recommended effect show a downward trend, and existing low recommendation precision, low recommendation accuracy and other issues.To solve the above problems, sentiment analysis technology combined with recommendation technology will be used in this thesis, using the product comment texts on the Internet, mining product characteristics which users are interested in, as well as the characteristics of the preference information, and then combined with the collaborative filtering recommendation algorithm to form a improvement algorithm that is based on collaborative filtering recommendation algorithm, which makes the recommendation results more suitable with the needs of users. In this thesis, the research content mainly includes the following aspects.Firstly, discusses the basic principle of recommendation algorithm, advantages and disadvantages, application fields of each algorithm, analyzing the current methods of combining the sentiment analysis technology with the recommendation system, and the exiting problems of these methods.Secondly, this thesis implement a personalized recommendation based on using the result of sentiment analysis to product reviews, this method is a improvement of the traditional collaborative filtering recommendation algorithm. To begin with, using natural language processing techniques to extract the feature and sentiment words from product comments, use graph database to structure data, and implement a method of sentiment analysis with product reviews based on graph database. And then, mining the product reviews information to find product characteristics which users are interested in. At last, computing the products similarity and the user similarity, and recommend the product with high evaluation characteristics to the other users with similar preferences.Finally, according to combine the above two jobs and the proposed mechanism to design recommendation system, using IMDB movie reviews data set to conduct an experiment. The results show that the system can effective predict the user of the product rating, it has a good recommend effect. Therefore, the method proposed in this thesis has a certain reference value, it is a feasible improvement scheme to the currently personalized recommendation system, which has a wide application prospect.