From literature to knowledge: Exploiting pubmed to answer biomedical questions in natural language

Researchers, practitioners and the general public strive to be constantly up to date with the latest developments in the subjects of biomedical research of their interest. Meanwhile the collection of high quality research papers freely available on the Web has increase dramatically in the last few years and this trend is likely to continue. This state of facts brings about opportunities as well as challenges for the construction of effective web-based searching tools. Question/Answering systems based on user interactions in Natural Language have emerged as a promising alternative to traditional keyword based search engines. However this technology still needs to mature in order to fulfill its promises. In this paper we present and test a new graph-based proof-of-concept paradigm for processing the knowledge base and the user queries expressed in natural Language. The user query is mapped as a subgraph matching problem onto the internal graph representation, and thus can handle efficiently also partial matches. Preliminary user-based output quality measurements confirm the viability of our method.