Domain-Specific Online Topic Modeling And Event Detection

In recent years, the Internet has produced lots and lots of information every day with the rapid development of technology. Hot topics and events especially which are closely related to people’s life were often triggered, transferred and diffused through the network first, and thus impose significant impact on social public security. On one hand, people express feelings and comment on current events more and more than before. On the other hand, false information will be propagated if we do not be careful about it and thus will cause great damage. So how to capture hot topics and events fast and efficiently in real-time through the streaming data has become a very important research aspect on public sentiment and social public security.As the requirement given above, this paper build a framework for topic modeling and detecting events in real-time on news stream. After that, we integrated related algorithm on it and then show our effort on topic modeling. Generally speaking, it has solved problems such as:1) how to find topics under the situation of big data and streaming data;2) how to detect events fast and effectively and extract features of these events. By our modular and data structure design, we can achieve good result promptly;3) how to carry out related topic mining work such as evolution and transfer and other issues. we can easily get answers by our algorithms and our expression of topics and events.After all, this paper aims for social public security and doing experiments on data set of food-security. The result shows that our work has good accuracy and speed, and the result is easy to explain. thus we say that it can meet our application requirements.