DYME: A Dynamic Metric for Dialog Modeling Learned from Human Conversations

Florian von Unold, Monika Wintergerst, Lenz Belzner, Georg Groh

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

1 Scopus citations

Abstract

With increasing capabilities of dialog generation methods, modeling human conversation characteristics to steer the dialog generation towards natural, human-like interactions has garnered research interest. So far, dialogs have mostly been modeled with developer-defined, static metrics. This work shows that metrics change within individual conversations and differ between conversations, illustrating the need for flexible metrics to model human dialogs. We propose DYME, a DYnamic MEtric for dialog modeling learned from human conversational data with a neural-network-based approach. DYME outperforms a moving average baseline in predicting the metrics for the next utterance of a given conversation by about 20%, demonstrating the ability of this new approach to model dynamic human communication characteristics.

Original languageEnglish
Title of host publicationNeural Information Processing - 28th International Conference, ICONIP 2021, Proceedings
EditorsTeddy Mantoro, Minho Lee, Media Anugerah Ayu, Kok Wai Wong, Achmad Nizar Hidayanto
PublisherSpringer Science and Business Media Deutschland GmbH
Pages257-264
Number of pages8
ISBN (Print)9783030923068
DOIs
StatePublished - 2021
Event28th International Conference on Neural Information Processing, ICONIP 2021 - Virtual, Online
Duration: 8 Dec 202112 Dec 2021

Publication series

NameCommunications in Computer and Information Science
Volume1516 CCIS
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937

Conference

Conference28th International Conference on Neural Information Processing, ICONIP 2021
CityVirtual, Online
Period8/12/2112/12/21

Keywords

  • Conversational metrics
  • Dialog modeling
  • Dialog systems
  • Natural language processing

Fingerprint

Dive into the research topics of 'DYME: A Dynamic Metric for Dialog Modeling Learned from Human Conversations'. Together they form a unique fingerprint.

Cite this