A Mixed Information Source Approach for Biomedical Question Answering - MindLab at BioASQ 7B.

This paper describes the participation of the MindLab research group in the BioASQ 2019 Challenge for task 7b, document retrieval and snippet retrieval. For document retrieval, Elastic Search was used for the initial document retrieval step with BM25 as a scoring function. In the second stage, the top 100 retrieved documents were re-ranked with several strategies to exploit embedding semantic similarity. For the snippets retrieval subtask, the proposed approach was based on textual and conceptual information similarity patterns that were combined into a feature matrix that was subsequently processed by a convolutional neural network architecture. Our approach reached the third and second positions for the document retrieval and snippet retrieval task respectively.