SU-NLP at TREC NEWS 2020

This paper presents our work and submissions for the TREC 2020 News Track: background linking and wikification tasks. For the background linking task, we investigated utilization of transformer-based architectures as either to map news articles to sentence embeddings, or to extract summaries of queried news articles or to re-rank the candidate documents. Even though none of these approaches outperformed our regular full-text baseline search approach, they provided some insights and future research directions. In the wikification task, after extracting and linking the entities to their corresponding entities, we analyzed two ranking approaches; one depends on entities positions within the text and the other depends on vector similarities between the news article and entity’s linked page.