ContReviews: A content-based recommendation system for updating Living Evidences in health care

Systematic reviews (SRs) summarise the knowledge available in the literature related to a specific research topic. Keeping SRs up-to-date with new publications as soon as they become available is fundamental to avoid their early obsolescence. Recently, automated methods have been proposed to update one or more SRs. However, particularly in the health care domain, it is necessary to scale these methods to maintain Living Evidences, which comprise thousands of SRs. In this context, the main issue of using the current methods is that they are SR-specific, that is, they require manually designing and optimising search queries and eligibility assessment models for each SR. To address this challenge, the ContReviews system is proposed. ContReviews first leverages an academic knowledge graph to gather new publications, and then uses a content-based recommendation model to match these new publications to all the SRs in a Living Evidence. To faithfully represent new publications and SRs, multiple publication properties are used (i.e., title, abstract, citation network, and authors) and, for each of them, likelihoods of relevance are calculated and used to learn a relevance assessment function for an entire Living Evidence. ContReviews has been evaluated on a dataset of 6000+ Cochrane Reviews in the health care domain, reporting high efficiency and high effectiveness in recommending new publications to the Cochrane Reviews. Specifically, ContReviews has achieved an average precision of 98.1% with a recall of 100% on the considered Cochrane Reviews.