In this paper, we present a system called Vise for visualizing salient events in a text stream according to the users' interests. A text stream is a sequence of chronological ordered documents. News articles, email and newsgroup postings are some typical examples of text stream. Through Vise, a user can visualize the events resides in a text stream by providing a set of keywords that are related to the events. A graph will be displayed to denote for the underlying patterns of the events. Yet, retrieving events in a text stream is a very difficult task due to the sparsity and noisiness of the features (keywords) in there. We solve these problems with the help of binomial distribution and some statis- tical theories. We have archived a stream of two-year news articles to evaluate the usability and the effec- tiveness of Vise. According to a subjective evaluation, the patterns of the events identified are justifiable and match our expectation. These favorable results indi- cated that our proposed system is highly effective and practical.
展开▼
