Research On Collaborative Filtering Algorithm Based On User Reviews And Ratings

In recent years,collaborative filtering algorithms have been widely concerned and applied in the field of recommendation.Traditional collaborative filtering algorithm uses user ratings for recommendation,however,with the increasing number of users and commodities,the sparse problem of rating data has become an important factor that restricts the recommendation effect of traditional collaborative filtering algorithm.Among the existing collaborative filtering algorithms,the fusion of review mining and collaborative filtering is one of the most important ways to alleviate the problem.The collaborative filtering algorithm based on topic model for review mining has attracted more and more attention because of its advantages such as mathematical statistics and flexible development of topic models.However,the existing algorithms do not take full account of the short text characteristics and emotional characteristics of user reviews.The accuracy of topic probability distribution based on topic model is difficult to be guaranteed,which has become an important factor restricting the recommendation effect of these algorithms.The pre filling method of the rating matrix is also one of the important ways to alleviate the data sparsity problem in the existing collaborative filtering algorithms.But the existing algorithms have different limitations in the rationality,applicability and accuracy of the filling methods,which can not effectively improve the quality of recommendation.In order to alleviate the effect of data sparsity on the effect of recommendation of existing collaborative filtering algorithms,this paper proposes an improved collaborative filtering algorithm based on user reviews and ratings from two aspects of review mining and rating matrix filling.First of all,in view of the limitations of the existing collaborative filtering algorithms in the use of topic models for review mining,this paper proposes a method of review mining based on the topic and sentiment hybrid model in collaborative filtering algorithm.The user sentiment-topic distribution obtained from the review mining is used to improve the similarity computation method of the existing collaborative filtering algorithm,which is conducive to obtaining more accurate similarity and improve the recommendation quality.Secondly,aiming at the limitations of the existing collaborative filtering algorithms for rating matrix filling,this paper proposes a method of filling the rating matrix using user sentiment-topic distribution and user interaction data.User interaction data is a new type of data generated in recent years in the development of e-commerce web site.It can reflect users' opinions to a certain extent,which is a valuable data resource.Finally,according to the data requirements of the algorithm,this paper uses the Python language to develop the crawler and crawls the commodity data on the Jingdong web site.This paper determines the configuration of relevant parameters through experiments,and compares the algorithm proposed in this paper with several existing algorithms.Experiment shows that the proposed algorithm can effectively improve the quality of recommendation.The research content of this paper further enriches the theory of personalized recommendation on the basis of existing research,and provides theoretical support and scientific basis for collaborative filtering recommendation algorithm research.