Research On Recommendation System Based On Reviews And Ratings Data

With the rapid development of information technology,the scale of data is getting larger and larger,and mankind has truly entered the era of big data.While this era provides us with great convenience,it also brings about the problem of "information overload".In order to solve this problem,researchers have proposed many solutions,among which the recommendation system is an important method to solve this problem.It can connect different users and projects,not only can the project be highly recommended to users,but also make it easy for users to find items of interest to them.However,in the recommendation system,the data sparsity caused by the lack of historical score data has always restricted the development of the recommendation system.As e-commerce technology matures and the platform interacts with users more and more,users begin to comment on items or services.These comments reflect the user's personalized needs and user's preference for items more than ratings.Therefore,this paper proposes a research method for recommendation systems based on reviews and ratings Data.Specifically,it can be divided into the following two tasks.A social matrix decomposition model ReTOMF based on reviews data is proposed to improve the sparsity problem caused by insufficient score data.The lack of historical scoring data can cause cold start problems,resulting in poor quality recommendations.This article uses rich reviews data,uses topic models to model documents,mines implicit neighbor relationships between users and projects,and integrates into social recommendation frameworks.In the past recommendation system,projects remained independent,but there should also be corresponding links between projects,so this article also establishes a social relationship between projects.When users buy unfamiliar items,they will choose to view reviews and other information to judge the value of the item,so add item reputation to the model.A recommendation model ReTFGM based on the sentiment features of reviews data is proposed to mine the sentiment features of users on items in the reviews data.First,use the topic model to calculate the topic distribution of each document,and use the score data to establish user comment attitudes,because the score data can be regarded as the numerical expression of the sentiment of the comment text,use the user comment attitude to improve the topic distribution,and form a new user preference is used to calculate the similarity between users and is called trust.Secondly,in order to make up for the problem of low trust caused by insufficient common reviews data,the concept of user reputation is proposed.The similarity between the user and its friend and the user's rating deviation are used to calculate the user's reputation,and as part of the calculation of the trust value,through trust The size of the value is for the target user to find the neighbor set and integrate it into the matrix decomposition technique based on trust propagation.Finally,in order to better improve the model and improve the accuracy of the recommendation system,the personalized characteristics of the user and the item,that is,the user's rating preference and item credibility,are added to the model.