Bridging The Evolving Semantics: A Data Driven Approach to Knowledge Discovery In Biomedicine

Recent progress in biological, medical and health-care technologies, and innovations in wearable sensors provide us with unprecedented opportunities to accumulate massive data to understand disease prognosis and develop personalized treatments and interventions. These massive data supplemented with rapid growth in computing infrastructure has enabled bio-medical researchers to perform more comprehensive experiments and detailed case-studies. At the same time, performing these experiments are not only monetarily expensive but also time consuming. Thus, there is a growing need to provide tools to the researchers that will allow them to pose queries that will assist them in focusing on interesting "hypotheses". However, such a tool would require capabilities to derive inferences based on existing known relationship between medical concepts. In this paper, we tackle this problem as one of non-factoid question answering wherein we try to answer the user-post questions by leveraging both authoritative as well as social media posts. While the former provides us with well knowledge on well researched topics, the latter provides us with real-time feedback on variety of topics like adverse drug effect (ADE), symptoms-drug relationship, etc. The challenge with leveraging the authoritative sources to infer answers for non-factoid question lies in: (a) The effective navigation of the answer search-space for timely response to the queries, (b) Ranking the candidate answers derived in step-(a) to enable non-trivial and novel discoveries, and (c) Being robust to perform confirmatory as well as discovery type of tasks.