Fight Against COVID-19 Misinformation via Clustering-Based Subset Selection Fusion Methods

ABSTRACT

The worldwide COVID-19 pandemic has brought about a lot of changes in people's life. It also emerges as a new challenge to information search services. This is because up to now our understanding about the virus is still limited, and there is a lot of misinformation online. In such a situation, how to provide useful and correct information to the public is not straightforward. Responsibility of search engines is crucial because many people make decisions based on the information available to them. In this piece of work, we try to improve retrieval quality via the data fusion technique. Especially, a clustering-based approach is proposed for selecting a subset of systems from all available ones for finding relevant, credible, and correct documents. Experimented with a group of runs submitted to the 2020 TREC Health Misinformation Track, we demonstrate that data fusion is a very beneficial approach for this task, whether measured by some traditional metrics such as MAP or some task specific metrics such as CAM. When choosing 17 runs, which is one third of all component retrieval systems available, the linear combination method is better than the best component retrieval system by 31.42% in MAP and 21.72% in CAM. The proposed methods are also better than the state-of-the-art subset selection method by a clear margin. © 2022 Copyright @Anonymous for this paper by its authors.