| Summary: | 碩士 === 國立臺北科技大學 === 資訊工程系 === 106 === With the popularity of social network, people use all kinds of social network platforms to express their opinion and participate in every kind of discussions. The traditional way of information propagation, like word-of-mouth or traditional media, change into propagating on social network. Furthermore, user can influence more people no matter the geo distance between them in the real world.
In order to quantify each person’s influence on social network platform. Most related works base on friend relationship or follow relationship and use different article attribute extraction to construct the social network for research and analysis. But it’s hard to perform without the friend or follow relationship in the network. In our research, we regard keyword as an interesting topic. According to the theory of information propagation, we build an influence model basing on the people who have the same interest and address their information and discussion in the past on the platform. First, in the same topic, we collect the relationship between articles and comments. According to the characteristic of information diffusion, we use adjacency list to construct the user relationship. Second, using the article relationship, we can get every user’s characteristic, like reply-count, degree, sentiment of comment. In the last, we extract every attribute, included evaluation of every posts, polarity of comments and articles relationship, to represent the influence in an efficient way. Moreover, we use the past influence to predict the new article’s user influence in the same topic in the future.
In this research, we focus on PTT social network platform. Using each topic influence model and their new issues as our experiment. According to the experiment results, our prediction model can achieve 92.9% in Average Precision. Compare to using centrality to estimate user influence, our model performs better in finding high influence user in the same topic.
|
|---|
