AUTHOR=Cai Yuanyuan , Zuo Min , Xiong Haitao 

TITLE=Modeling hierarchical attention interaction between contexts and triple-channel encoding networks for document-grounded dialog generation

JOURNAL=Frontiers in Physics

VOLUME=Volume 10 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2022.1019969

DOI=10.3389/fphy.2022.1019969

ISSN=2296-424X

ABSTRACT=Dialogue systems have attracted attention as it is promising in many intelligent applications. Generating fluent and informative responses is of critical importance for dialogue systems. Some recent studies introduce documents as extra knowledge to improve the performance of dialogue generation. However, it is hard to understand the unstructured document and extract crucial information related to dialogue history and current utterance. This leads to uninformative and inflexible responses in existing studies. To address this issue, we propose a generative model on neural network with attention mechanism for document grounded multi-turn dialogue. This model encodes the context of utterances that contains the given document, dialogue history and the last utterance into distributed representations via a triple-channel. Then it introduces a hierarchical attention interaction between dialogue contexts and previously generated utterance into the decoder for generating an appropriate response. We compare our model with various baselines on dataset CMU_DoG in terms of the evaluation criteria. The experimental results demonstrate the state-of-the-art performance of our model as compared to previous studies. Furthermore, the results of ablation experiments show the effectiveness of the hierarchical attention interaction as well as the triple-channel for encoding. We also conduct human judgment to evaluate the informativeness of responses and the consistency of responses with dialogue history.