Method Research of New Event Detection Based on News Time and Place Element
|Course||Applied Computer Technology|
|Keywords||News event detection Event model Calculation of similarity Time message Place message|
With the rapidly development of Internet and the enormous increase of computer users’, Internet has become an important means for obtaining information. However, useful information on the Internet is buried in a large amount of data that grows rapidly with time, The people are very difficult to gain the useful information from that complex network information resource. Especially in nowadays, emergency event has become a major problem that puzzled all over the world, people are more interested in the whole event and its influence on society, as well as the personals involved need to know theevent developments at the first time. The object of this study——New eventdetection is the sub-task of TDT, it is aimed at detecting from one or multiple streams of news stories.The paper is chiefly concerned with the effect of time and place element, we consider time and place element as an important measurement of the time similarity calculation, and make a deep analysis of the drift of the news cluster in NED. The updating scheme proposed by the present paper is aimed to decrease this influence of the drift. The main efforts in this paper are as follows:1. We take the way of feature weighted to improve the traditional term frequency inverse document frequency （TF-IDF） model, we introduce a forkωto adjust term weights, the term frequency is updated which is lower but more important, and then the new event detection reaches high precise.2. We propose the method research of news event detection based on news element and adopt time message and place message to enhance the efficiency of the new event detection. this paper extracts the message of time and place and normalizes, then calculates time similarity, place similarity and content similarity respectively ,finally combines their similarities for NED.3. We adopt the automatic updating scheme for the news cluster template, and then the template is updated dynamically with the news event increasing. 4. We propose five NED models to address the problem. Besides this, Miss Rate, False Alarm rate and CDet function are carried on to evaluate the results.