Training Data Optimization for Persona-grounded Dialog via Synthetic Label Augmentation

Anonymous

TL;DR: We aim to refine the Persona-Chat dataset to develop a more advanced conversational AI. To achieve this, we propose the Synthetic Label Augmentation framework, leveraging AI models to optimize the dataset

Abstract: The goal of persona-grounded dialogue systems is to enhance the quality of AI agent responses by bolstering persona consistency and promoting response diversity. Although model tuning has seen significant advancements, there is an ongoing need to refine the training data itself. Expanding the scope of personas has been suggested as a means to bridge this gap. Nevertheless, the lack of gold labels that align with these expanded personas poses a challenge for AI agents in training the extent of real-world knowledge. To tackle these challenges, we propose the Synthetic Label Augmentation framework. This framework (1) creates a background skeleton from the original gold labels, masking persona-related elements, (2) infuses the background skeleton with expanded-persona features, generating synthetic gold labels, (3) identifies the most appropriate synthetic gold labels among the candidates, and (4) merges them into Persona-Chat. To substantiate the effectiveness of Optimized Persona-Chat, we assess the quality of synthetic gold labels and interact with agents trained on this enhanced dataset. Our experimental results demonstrate that the framework is a powerful tool for augmenting Persona-Chat quality, and the optimized dataset significantly improves AI agent response quality with respect to persona consistency and response diversity.

Paper Type: long

Research Area: Dialogue and Interactive Systems

Contribution Types: NLP engineering experiment, Data resources, Data analysis

Languages Studied: English