Learning what to read: Focused machine reading

摘要

Recent efforts in bioinformatics have achieved tremendous progress in themachine reading of biomedical literature, and the assembly of the extractedbiochemical interactions into large-scale models such as protein signalingpathways. However, batch machine reading of literature at today's scale (PubMedalone indexes over 1 million papers per year) is unfeasible due to both costand processing overhead. In this work, we introduce a focused reading approachto guide the machine reading of biomedical literature towards what literatureshould be read to answer a biomedical query as efficiently as possible. Weintroduce a family of algorithms for focused reading, including an intuitive,strong baseline, and a second approach which uses a reinforcement learning (RL)framework that learns when to explore (widen the search) or exploit (narrowit). We demonstrate that the RL approach is capable of answering more queriesthan the baseline, while being more efficient, i.e., reading fewer documents.