LDA-TIM: An approach for individual sentiment prediction in social networks

Abstract

Social networks like Facebook and SINA have been rapidly growing and accumulating a sheer volume of data such as social links between the users, user claims, and their comments. The work is motivated by the proliferation of social networks and large amounts of information that is voluntarily broadcast on them, which generates an interest in finding ways to predict individual sentiment that applies in public sentiment warning, advertisement, and recommendation. However, the traditional user sentiment prediction model has shortcoming of high complexity, which renders inefficiencies of individual sentiment prediction in social networks. To tackle this challenge, in this study, we develop an individual sentiment prediction method LDA-TIM based on the individual interest preferences and social influence. Then, based on the objective function we trained a logistic regression classifier to predict individual sentiment polarity. Finally, extensive experiments are conducted to evaluate the performance of our approach by using two large-scale real-world data collected from SINA. The experimental results on the two large-scale-real-word data set both reveal that each of the components are critical to obtaining satisfactory performance on our data. Experiments show the F1-Measure value of the individual sentiment approach can reach 70.99%.