SynCPKL: Harnessing LLMs to Generate Synthetic Data for Commonsense Persona Knowledge Linking

TL;DR

The paper tackles grounding open-domain dialogues in persona-based commonsense knowledge by generating synthetic data with LLMs to train a commonsense persona knowledge linker. It introduces the SynCPKL pipeline and the SynCPKL dataset, built from PeaCoK and PersonaChat, and demonstrates that synthetic data can outperform a ComFact-based baseline. The top-performing Derberta-SynCPKL model achieves first place in the CPKL challenge, with a 16% F1 improvement, and the authors release both data and model to accelerate progress. The approach addresses data scarcity for CPKL, provides ablation-driven insights into head–tail integration, and highlights practical impact for improving grounding in open-domain dialogue systems.

Abstract

Understanding rich dialogues often requires NLP systems to access relevant commonsense persona knowledge, but retrieving this knowledge is challenging due to complex contexts and the implicit nature of commonsense. This paper presents our approach to the Commonsense Persona Knowledge Linking (CPKL) challenge, addressing the critical need for integrating persona and commonsense knowledge in open-domain dialogue systems. We introduce SynCPKL Pipeline, a pipeline that leverages Large Language Models to generate high-quality synthetic datasets for training commonsense persona knowledge linkers. To demonstrate the efficacy of our approach, we present SynCPKL, a new dataset specifically designed for this task. Our experiments validate the effectiveness of SynCPKL for training commonsense persona knowledge linkers. Additionally, our top-performing model, Derberta-SynCPKL, secured first place in the CPKL challenge by a 16% improvement in F1 score. We released both SynCPKL and Derberta-SynCPKL at https://github.com/irislin1006/CPKL.