The User Rating Prediction Research Based On Topic Model In Social Media

With the rapid development of Internet, people have been used to publishing their opinions, views and feelings on the Internet, as well as obtaining needed information from others. This situation leads to a new Internet mode based on extensive users. Social media is developed in such an environment. Social media is the virtual community or network platform where users can create, share and exchange their opinions, views and experiences. Users publish their opinions on the social media, and the personal opinions often contain abundant sentiments. These opinions can be divided into two types. One is the text information, such as articles in microblogs, and reviews in websites. The other is the rating information,such as ratings for products.User rating prediction research is the analysis and study of the information that contains abundant sentiments. The goal of user rating prediction is to predict the possible rating of a product for a given user.Because of the personalized needs for specific aspect evaluation on product quality, there is a surge of researches on aspect rating prediction,whose goal is to extract ad hoc aspects from online reviews and predict rating or opinion on each aspect. Therefore, this paper studied the user rating prediction problem in social media from the perspective of aspect rating. However, the aspect rating prediction problem faces two challenges: (1) How to effectively integrate rating and review information?(2) How to effectively integrate content and network structure information of objects? To solve these challenges, we designed two novel aspect rating prediction models which are MaToAsp and HINToAsp,respectively.How to effectively integrate rating and review information is the key issue for aspect rating prediction. Since matrix factorization is an effective tool for rating prediction and topic model is widely used for review processing, it is a natural idea to combine matrix factorization and topic model for aspect rating prediction. However, it faces several challenges: suitable sharing factors, scale mismatching, and dependent relation of rating and review information. In this paper, we propose a novel model to effectively integrate Matrix factorization and Topic model for Aspect rating prediction, called MaToAsp. The experiments on two real datasets including Chinese and English show that the MaToAsp not only generates reasonable aspect identification, but also achieves the best aspect rating prediction performance, compared to representative baselines.Recently, heterogeneous information network becomes a research hot spot in data mining field. Compared to homogeneous information network, it contains different types of nodes and links as well as more complex structure information. It also can convey richer semantic information. Topic model is an important method in text analysis field,which has wide applications in text mining and rating prediction.However, topic model only considers the text content of an object, and ignores the network structure. In this paper, we combine heterogeneous information network and topic model, and design a unified model HINToAsp to solve the aspect rating prediction problem. HINToAsp employs topic model to model the content information of an object, and employs heterogeneous information network to model the structure information. A biased random walk framework is designed to obtain mutual effect and enforcement between these two kinds of information,so that the accuracy of aspect rating prediction can be improved. The experiments on Chinese and English datasets validate the effectiveness of HINToAsp.