An End-to-end Question Answering Model Based on Semantic-enhancing Attention Mechanism

Abstract

Task-oriented question answering dialogue systems have been an important branch of conversational systems for oral language, where they first understand the query requested by users, and the models are demanded to seek for answers within the context considering the query information. Previous work models the semantic and syntactic information without taking the interaction into consideration. In this paper, we propose an end-to-end model based on semantic-enhancing attention mechanism, which enables the model to focus more on a small part of the context and enhances the model capability of extracting the interactive information. Our experiments are based on the Stanford Question Answering Dataset (SQuAD) and the experimental result verifies how the proposed model improves on the dataset.