Microblog Oriented Hot Topic Extraction And Its Sentiment Classification

With the advent of Web2.0era, especially the rapid development of social network,more and more internet users begin to publish their opinions towards some topics on theweb. How to detect the real-time hot topics and analyze users’ opinions towards themhave become a meaningful research point. In this paper, we study it in-depth.Microblog oriented hot topics are different from common topics. In Microblog, hottopics means the topics which attract a lot of users’ attention in a very short time. Thesehot topics can be real social event or can just have nothing to do with social event, butstill get a lot of users’ attention in a very short time.This paper divides the task into two subtasks: Microblog oriented hot topicdetection and its sentiment classification.For Microblog oriented hot topic detection task, we propose a hashtag based hottopic detection method. This method utilizes heuristic rules and clustering algorithm toextract real-time hot topics in Microblog. This algorithm mainly includes three parts:topic find, topic clustering and topic rank. Experiments show a high accuracy about thissubtask.For topic sentiment classification, we propose a discourse relation based sentimentclassification method. We sum up four kinds of discourse relations that may haveimpact on sentiment classification result, and then utilize their corresponding explicitconnectives to extract the trunk of microblog text. On this basis, we explore a variety ofsentiment classification features. Besides, we alse build a Microblog oriented sentimentlexicon, which has a very large scale. Experiments show that our method can obtain agood sentiment classification result.Besides, we merge the two algorithms described above and build a real-time useremotion analysis platform. This platform can detect real-time hot topics in Microblogand analyze users’ opinions towards them. This task is very meaningful.