Retrieval of Semantically Similar Healthcare Questions in Healthcare Forums

Healthcare forums are popular platforms for patients to communicate with other patients who have similar conditions. Retrieving similar post to a user´s question is valuable as the question might have been already answered in similar threads. ICHI 2015 organized a challenge with a corpus of ninety-five selected questions and a query set of ten questions from various diabetes related online forums. The task in this challenge is that given the corpus, a system should be developed to generate the most three similar questions for each question in the query set. In order to accomplish the challenge, we utilized Elastic search to built and search an index based on tokens, UMLS concepts, and semantic types, and finally combined the ranking results with LDI, a Latent Dirichlet Allocation (LDA) based ranking method. The experimental results showed that a mean average precision of 0.72 was achieved on the manually created gold standard.