EnDex: Evaluation of Dialogue Engagingness at Scale

• Main Authors: Xu, Guangxuan, Liu, Ruibo, Harel-Canada, Fabrice, Chandra, Nischal Reddy, Peng, Nanyun
• Format: Journal Article
• Language: English
• Published: 22.10.2022
• Subjects: Computer Science - Computation and Language

Findings of EMNLP 2022 We propose EnDex, the first human-reaction based model to evaluate dialogue engagingness. EnDex is trained on 80k Reddit-based Engagement Dataset (RED) curated using a novel distant-supervision framework. Engagingness is a key measure that captures high-level quality of AI dia...