| Topic detection and tracking (TDT) is a research which helps people to resolve the problem of overload information .TDT can auto-recognize the new topic and dynamically track the known topic from news stream, deal with the media information, such as newswire, broadcast, television and so on, divide the information of language into different news story. In addition, it can detect the story of new topics, organize stories about same topic and present it in a certain way to the customers. The research goal is to find, organize and use multi-language information from various news medium by topic. Topic tracking is the sub-task of TDT .This paper tracks the sequential story of accidental event, lets people know the latest evolution of the event and obtain all details and the relation between this event and other events.This paper researches vector space model based on key elements uses the topic tracking method based on query vector and tracks the sequential story of accidental news. Considering the importance of key elements (5H1W) of news story, we try to use temporal information, place information and event content to express news story, so as to track the sequential story of accidental events and improve the efficiency of tracking .Main work of this paper is as follows:1. We statistic the corpus of accidental news which downloads from internet and analyze the characteristics of accidental news story and the relation between accidental news story and its sequential story, so as to use temporal information, place information. At the same time, according to the regionalism of country, we establish the place-base, including province, city, zone, county, town and so on. In addition, we complement some interrelated concepts, such as event, story, sequential story and so on.2. We analyze the characteristics of using temporal information and place information, recognize those information by NE tool, extract and standardize temporal and place information of news story .For temporal information, the paper compares the time relation between the previous event and later story .For place information, we use the correlativity of place granularity, the location of place in news story and so on to calculate the similarity of place.3. We combine temporal information, place information and event contents to express news story and present a tracking algorithm based on key elements of accidental news, which uses spatio-temporal information to improve tracking efficiency.4. We design an experiment system. This system can extract place and temporal information, calculate similarity and track the sequential story of accidental news.In order to prove the validity of this method, we select 20 events from accidental news corpus, about 880 stories as our experimental corpus. The result shows that the method of this paper improves the tracking efficiency in a certain extent. |
PDF Full Text Request