Recommending research articles to consumers of online vaccination information

Abstract

Research communications often introduce biases or misrepresentations without providing reliable links to the research they use so readers can check the veracity of the claims being made. We tested the feasibility of a tool that can be used to automatically recommend research articles to research communications. From 207,538 vaccination-related PubMed articles, we selected 3,573 unique links to webpages using Altmetric. We tested a method for ranking research articles relative to each webpage using a canonical correlation analysis (CCA) approach. Outcome measures were the median rank of the correct source article; the percentage of webpages for which the source article was correctly ranked first; and the percentage ranked within the top 50 candidate articles. The best of the baseline approaches ranked the matching source article first for more a quarter of webpages; and within the top 50 for more than half. Augmenting baseline methods with CCA improved results but failed when added to some of the baseline approaches. The best CCA-based approach ranked the matching source articles first for 14%; and in the top 50 for 38%. Tools to help people identify source articles for vaccination-related research communications are potentially feasible and may support the prevention of bias and misrepresentation of research in news and social media.