HisTrace: A system for mining on news-related articles instead of web pages

The Web is now playing an important part in people's real-life activities. Scientists of not only computer science but also sociology and economics might be interested in mining on information directly related to real-life events, or news-related information on the Web. In this paper we propose a system to enable mining on news-related articles instead of raw web pages. There are functionally two tasks in our system: 1) mining for news-related articles and 2) duplicate elimination. For the first task, a novel approach for determining titles, contents and publication-times of news-related articles is presented. Anchor texts are firstly used to extract titles from HTML bodies and then contents are extracted right after titles. After that, crawl-times and are used to initially compute publication-times for all articles. At last, times extracted from HTML bodies, URLs and anchor texts are used to determine precise publication-times for possible articles. For the second task, a duplicate detection algorithm for news-related articles is described which is base on LCS (longest common subsequence) and achieves both high precision and high recall. The framework of this algorithm has been presented as a general-purpose algorithm for web pages in a previously published paper. In this paper we explain why this algorithm is particularly suitable for news-related articles and present corresponding implementation details. Evaluations have been conducted which show the effectiveness of our approaches.