Please use this identifier to cite or link to this item:
|Title:||Exploiting User Interaction Network for User-level Sentiment Analysis in Weibo|
|Department:||Department of Computer Science|
|Supervisor:||Dr. Li, Shuai Cheng|
|Abstract:||With the fast development of microblog platforms, increasing number of people tend to share their feelings and opinions on microblog website, which makes microblogging websites a rich resource for sentiment analysis. However, there are two problems when performing sentiment analysis on microblog data. The first problem is that although Chinses microblog platforms like Sina Weibo is under fast development, only a few research works were devoted to the sentiment analysis of Chinese Microblog data. The second problem is that most sentiment analysis tools focus only on the content and fail take user interactions between microblog users into consideration, So, in this paper, I will focus on Weibo data, and propose a novel approach to compute user-level sentiment status by exploiting user interaction information. At beginning, I have introduced three off-the-shelf sentence-level sentiment analysis tools and implemented an original sentence-level sentiment analysis method which exploits emoticons in Weibo messages as a feature for sentiment classification. A consensus sentiment prediction model with relatively high accurate rate can be constructed by combining the result of these four sentiment analysis approaches. But this is not the end of the story. Then we need to exploit network information into the result. To do this, I imagine that the microblog is a huge graph, where each node represents a user and each edge indicates user interaction. Every user in this graph is able to influence other people and get influence by others through user interaction at the same time. So, for each node, the total sentiment influence that all of its neighbors offer to it can be regarded as the information carried by user interaction network. On the base of this network feature and sentiment prediction results of other four sentiment analysis tools, we could definitely improve the correct rate into a new level.|
|Appears in Collections:||Computer Science - Undergraduate Final Year Projects |
Items in Digital CityU Collections are protected by copyright, with all rights reserved, unless otherwise indicated.