Skip to main navigation Skip to search Skip to main content

Training Two-Stage Knowledge-Grounded Dialogues with Attention Feedback

  • Zhen Li
  • , Jiazhan Feng
  • , Chongyang Tao
  • , Dongyan Zhao*
  • *Corresponding author for this work
  • Peking University
  • Microsoft USA
  • State Key Laboratory of Media Convergence Production Technology and Systems

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Knowledge-grounded retrieval-based dialogue systems have attracted more and more attention. Among them, the two-stage dialogue models which separate the training stage into knowledge retrieving (via a retriever) and response ranking (via a ranker) are proved powerful. However, these approaches require knowledge-grounded dialogues with corresponding hand-annotated knowledge labels. Therefore, in this paper, we propose training two-stage knowledge-grounded dialogues with knowledge attention feedback from the ranker to the retriever. In each training iteration, the ranker provides knowledge attention scores as pseudo supervised feedback for the optimization of retriever. We conduct experiments on two public data sets. The experimental results demonstrate that our proposed method is superior to the existing baselines.

Original languageEnglish
Title of host publicationNatural Language Processing and Chinese Computing - 11th CCF International Conference, NLPCC 2022, Proceedings
EditorsWei Lu, Shujian Huang, Yu Hong, Xiabing Zhou
PublisherSpringer Science and Business Media Deutschland GmbH
Pages473-484
Number of pages12
ISBN (Print)9783031171192
DOIs
StatePublished - 2022
Externally publishedYes
Event11th CCF International Conference on Natural Language Processing and Chinese Computing, NLPCC 2022 - Guilin, China
Duration: 24 Sep 202225 Sep 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13551 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference11th CCF International Conference on Natural Language Processing and Chinese Computing, NLPCC 2022
Country/TerritoryChina
CityGuilin
Period24/09/2225/09/22

Keywords

  • Knowledge attention feedback
  • Knowledge-grounded dialogues
  • Multi-turn context modeling
  • Retrieval-based dialogues

Fingerprint

Dive into the research topics of 'Training Two-Stage Knowledge-Grounded Dialogues with Attention Feedback'. Together they form a unique fingerprint.

Cite this