Finding answers to COVID-19-specific questions: An information retrieval system based on latent keywords and adapted TF-IDF

Abstract

The scientific community has reacted to the COVID-19 outbreak by producing a high number of literary works that are helping us to understand a variety of topics related to the pandemic from different perspectives. Dealing with this large amount of information can be challenging, especially when researchers need to find answers to complex questions about specific topics. We present an Information Retrieval System that uses latent information to select relevant works related to specific concepts. By applying Latent Dirichlet Allocation (LDA) models to documents, we can identify key concepts related to a specific query and a corpus. Our method is iterative in that, from an initial input query defined by the user, the original query is expanded for each subsequent iteration. In addition, our method is able to work with a limited amount of information per article. We have tested the performance of our proposal using human validation and two evaluation strategies, achieving good results in both of them. Concerning the first strategy, we performed two surveys to determine the performance of our model. For all the categories that were studied, precision was always greater than 0.6, while accuracy was always greater than 0.8. The second strategy also showed good results, achieving a precision of 1.0 for one category and scoring over 0.7 points overall.

Comparing COVID-19 vaccine passports attitudes across countries by analysing Reddit comments

“…By using text mining technique with LDA procedure to undertake an overview of a large body of coronavirus literature, Cheng et al [ 14 ] showed how information specialists could benefit the health and medical community. Chamorro-Padial et al [ 15 ] described an information retrieval system that chooses pertinent works associated with particular concepts using latent information. They found important ideas connected to a certain query and a corpus by using LDA models on COVID-19-related articles.…”