bigIR at TREC 2019: Graph-based Analysis for News Background Linking

Abstract

Nowadays, it is very rare to find an online news article that is self-contained with everything a reader would want to know about the article's story. Therefore, it became vital for any article to contain links to other articles or resources that provide the background and contextual knowledge required to conceptualize the article's story. However, finding useful background and contextual links can be a challenging problem. In this paper, we address this problem in the context of the participation of the bigIR team at Qatar University in the news background linking task of the TREC 2019 news track. Our methods mainly relied on a graph-based analysis of the query-article's text to extract its most representative and influential keywords, and then use these keywords as a search query to retrieve the article's background links from a collection of news articles. All of our submitted runs outperformed the TREC hypothetical run that achieved a median effectiveness over all queries. Moreover, our best submitted run was ranked second among 28 runs submitted to the task, indicating the potential effectiveness of our approach.