Dynamic Cluster-based Retrieval and Discovery for Biomedical Literature

Abstract

Due to increased specialization and experimentation, the volume of biomedical literature is rapidly increasing, where the current modalities of search and retrieval system can no longer support effective and efficient knowledge discovery. Standard information retrieval systems such as PubMed make assumptions as to users' prior knowledge and expect them to formulate a proper query term for the information they are looking for. There exist user feedback mechanisms to help users reformulate their queries, which still assumes that users know how the search results could be effectively narrowed down by way of additional keywords and/or filters. As an alternative, we revisit the Scatter/Gather information retrieval paradigm. Specifically, we explore a real-time dynamic cluster-based document browsing approach in the biomedical domain, discuss the system architecture involving keyword discovery and dynamic clustering, and present a working prototype with a relevant use case in comparison with a standard ranking-based information retrieval system.