Background: Healthcare consumers are increasingly turning to the online health Q&A communities to seek answers for their questions because current general search engines are unable to digest complex health-related questions. Q&A communities are platforms where users ask unstructured questions from different healthcare topics.
Objectives: This study aimed to provide a concept-based approach to automatically assign health questions to the appropriate domain experts.
Methods: We developed three processes for (1) expert profiling, (2) question analysis and (3) similarity calculation and assignment. Semantic weight of concepts combined with TF-IDF weighting comprised vectors of concepts as expert profiles. Subsequently, the similarity between submitted questions and expert profiles was calculated to find a relevant expert.
Results: We randomly selected 345 questions posted by consumers for 38 experts in 13 health topics from NetWellness as input data. Our results showed the precision and recall of our proposed method for the studied topics were between 63 %-92 % and 61 %-100 %, respectively. The calculated F-measure in selected topics was between 62 % (Addiction and Substance Abuse) and 94 % (Eye and Vision Care) with a combined F-measure of 80 %.
Conclusions: Concept-based methods using unified medical language system and natural language processing techniques could automatically assign actual health questions in different topics to the relevant domain experts with good performance metrics.
Keywords: Concept-based similarity; Expert profiling; Question answering systems; Similarity measures; UMLS.
Copyright © 2020 Elsevier B.V. All rights reserved.