Towards a Two-Stage Method for Answer Selection and Summarization in Buddhism Community Question Answering

Jiangnan Du, Jun Chen, Suhong Wang, Jianfeng Li, Zhifeng Xiao

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

This paper proposes a two-stage learning pipeline for CQA in the Buddhism domain. In the first stage, we trained an answer selection model through Keywords-BERT that performs a deep semantic match for QA pairs. Given a question, our algorithm selects the answer with the highest relatedness score. Stage two also employs the trained Keywords-BERT model to eliminate redundant information and only keep the most relevant sentences of an answer for summary extraction. Our method only requires standard QA pairs for training, significantly reducing the annotation cost and the knowledge threshold for annotators. We tested our model on a self-created Buddhism CQA dataset. Results show that the proposed pipeline outperforms state-of-the-art methods like BERT-Sum in terms of summary quality and model robustness.

Original languageEnglish (US)
Title of host publicationArtificial Intelligence - 1st CAAI International Conference, CICAI 2021, Proceedings
EditorsLu Fang, Yiran Chen, Guangtao Zhai, Jane Wang, Ruiping Wang, Weisheng Dong
PublisherSpringer Science and Business Media Deutschland GmbH
Pages251-260
Number of pages10
ISBN (Print)9783030930486
DOIs
StatePublished - 2021
Event1st CAAI International Conference on Artificial Intelligence, CICAI 2021 - Hangzhou, China
Duration: Jun 5 2021Jun 6 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13070 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference1st CAAI International Conference on Artificial Intelligence, CICAI 2021
Country/TerritoryChina
CityHangzhou
Period6/5/216/6/21

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint

Dive into the research topics of 'Towards a Two-Stage Method for Answer Selection and Summarization in Buddhism Community Question Answering'. Together they form a unique fingerprint.

Cite this