Toward a Deep Multimodal Interactive Query Expansion for Healthcare Information Retrieval Effectiveness

The emerging trend of Health Information Retrieval (HIR) aims to efficiently address users’ specific information requirements. However, a beyond challenge arises in the healthcare domain due to the necessity for specialized dictionaries, which influence the outcomes of HIR. The Vocabulary Mismatch (VM) phenomenon necessitates more robust efforts in the Health Information Retrieval domain, mainly in query formulation. It is crucial to support clinicians, biomedical scientists, and non-specialists in their daily retrieval endeavors. In this paper, we propose an innovative approach that combines deep image captioning for clinical diagnosis generation then MetaMap normalization, and finally a user involvement step. This combination aims to optimize query formulation task for an enhanced query and an efficient HIR process. Experimental results, conducted on widely used search engines such as Google and Bing, reveal that our approach has demonstrated its effectiveness by enhancing result quality and delivering documents from reliable sources. It has significantly improved the user experience, ensuring relevant results appear within the top five ranking links. The findings demonstrate that the integration of various techniques is a valuable enhancement to the Query Expansion (QE) process, resulting in a notable increase in weighted precision rates by around the twofold a MAP to over 70% and an apparent reduction in Vocabulary Mismatch for Healthcare Information Retrieval.