Research On Extraction And Tracking Of People's Opinion

With the rapid explosion of information coming from internet, people are in the very need of tools for automatic acquisition of information, which can help them to find meaningful and useful information from the web. This paper does an intensive research on extraction and tracking against people's opinion, with the aim of finding and extracting a cluster of opinions, which focus on the same topic as interested in by the user, of people or a group or an organization etc.This paper is focusing on two key issues of the research, one of which is named recognition and extraction of people's opinion, and the other is named tracking on people's opinion related to a specific topic.This paper regards the appearance of certain people's opinion as a specific kind of event– opinion event. In the research we have established a well-scaled opinion event corpus for the purpose of training and evaluation. This paper describes event as several features, using MEN based binary classification to recognize the true opinion events from candidate events. Experiment results shows the best performance of F value as 87.8 when using the optimized trigger set and using a training corpus including 350 news stories, 716 positive and 160 negative.This paper is using a tracking technique based on query vector. Comparing the separate ways of calculating similarity between two vectors, we obtain the best performance of norm DET as 0.3908, while using query vector extension and adjusting the threshold.