A Generic Framework for Biomedical Snippet Retrieval

In the biomedicine domain, a large number of papers are published every day, which is crucial to search for the relevant answers for the user's query. However, documents are not exactly what the users want. Instead, snippets, small segments from the documents, are more proper to meet the requirement of the users. Hence, this paper proposes a biomedical snippet retrieval framework to exactly locate the answers (snippets) for the biomedical questions (queries) from the users. In our framework, word embeddings with biomedical meanings are trained for query expansion. Then, the expansion query is used to retrieve the relevant snippets from the candidate documents using the sequential dependence model which can benefit retrieval results. Finally, our proposed framework is evaluated on the BioASQ 2014 task datasets, and has the best performance (MAP@100) compared to state-of-the-art systems. Moreover, our framework has competitive results on the BioASQ 2015 task.