CHIME is a cross-passage hierarchical memory network for generative question answering (QA). It extends XLNet introducing an auxiliary memory module consisting of two components: the context memory collecting cross-passage evidences, and the answer memory working as a buffer continually refining the generated answers. Following is samples of syntactically well-formed answers show the efficacy of CHIME.
- Task: We focus on generative QA with multiple reviews and develop our model based on the AmazonQA Dataset in which most of the questions is paired with multiple answers and the top10 most relevant text snippets as supporting passages extracted from the associated reviews by BM25. In addition, each question is annotated if it is answerable based onthe top 10 review snippets, and each answer is accompanied with response votes. The goal is to generate an answer response given the question and reviews. Next is the illustration of the task:
- Architecture: The overall architecture of CHIME is shown in the following figure. Given a question paired with K text passages, we create K training instances with each one consisting of the question, a text passage, and the best answer chosen by the helpfulness votes assigned by users. Each training instance is fed into an XLNet encoder to derive hidden representations, which will be used to update two memories. In particular, the context memory is updated when seeing more text passages and the answer memory is continuously refined with the answer generated from each (question, text passage) pair.
Junru Lu, Gabriele Pergola, Lin Gui, Binyang Li and Yulan He. CHIME: Cross-passage Hierarchical Memory Network for Generative Review Question Answering. In Proceedings of the 28th International Conference on Computational Linguistics (COLING), 2020.