Context-guided Triple Matching for Multiple Choice Question Answering

The task of multiple choice question answering (MCQA) is to identify the correct answer from multiple candidates given a passage and a question. It is typically approached by estimating the matching score among the triple of the passage, question and candidate answers. Existing methods decouple this estimation into pairwise or dual matching ignoring the third component. This paper introduces a Context-guided Triple Matching algorithm, which models the matching among the triple simultaneously. Specifically, the proposed matching takes one component from the triple as the context, and estimates its semantic matching between the other two. Additionally, a contrastive term is adopted to model the dissimilarity between the correct answer and distractive ones. The proposed algorithm is validated on several benchmarking MCQA datasets and outperforms the state-of-the-art models by a large margin.