In this paper, we present PersonaGen, a novel emotion-rich text generation framework using large-scale language models (LLMs) to address the lack of high-quality and diverse emotion datasets in the field of emotion recognition. PersonaGen constructs hierarchical virtual personas by combining demographic attributes, sociocultural backgrounds, and detailed situational contexts to drive emotion expression generation. We perform comprehensive evaluations, including semantic diversity assessment via clustering and distribution metrics, human-likeness assessment via LLM-based quality scores, realism assessment via comparison with real emotion corpora, and practicality assessment for downstream emotion classification tasks. Experimental results show that PersonaGen significantly outperforms baseline methods in generating diverse, consistent, and discriminative emotion expressions, demonstrating its potential as a powerful alternative to complement or replace real emotion datasets.
