SYNTHIA: Synthetic Yet Naturally Tailored Human-Inspired PersonAs

2025-07-20

AI Summary

Existing approaches to persona modeling face a dichotomy: high-cost manual annotation versus synthetic personas lacking authenticity and consistency. This paper introduces the first temporal, persona-grounded narrative dataset integrating real user behavior—comprising 30,000 background narratives derived from multi-temporal activity traces of 10,000 users on the Bluesky platform. It uniquely incorporates temporal dynamics and social interaction metadata, synergizing synthetic data generation with social network analysis. Evaluated across demographic diversity, narrative coherence, and alignment with sociological survey benchmarks, the dataset significantly outperforms state-of-the-art alternatives. It bridges the gap between labor-intensive human annotation and purely synthetic data, establishing a new benchmark for computational social science and persona-driven large language models—one that simultaneously ensures empirical fidelity, logical consistency, and scalable construction.

Technology Category

• Application Domains: Humanities & Computational Social Science
• Humans and AI: Game Design — Procedural Content Generation & Storytelling
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization

Abstract

Persona-driven LLMs have emerged as powerful tools in computational social science, yet existing approaches fall at opposite extremes, either relying on costly human-curated data or producing synthetic personas that lack consistency and realism. We introduce SYNTHIA, a dataset of 30,000 backstories derived from 10,000 real social media users from BlueSky open platform across three time windows, bridging this spectrum by grounding synthetic generation in authentic user activity. Our evaluation demonstrates that SYNTHIA achieves competitive performance with state-of-the-art methods in demographic diversity and social survey alignment while significantly outperforming them in narrative consistency. Uniquely, SYNTHIA incorporates temporal dimensionality and provides rich social interaction metadata from the underlying network, enabling new research directions in computational social science and persona-driven language modeling.

Problem

• Bridging synthetic and human-curated persona data gaps
• Enhancing persona consistency and realism in LLMs
• Incorporating temporal and social metadata for research

Innovation

• Synthetic personas grounded in real user activity
• Incorporates temporal dimensionality and metadata
• Enhances narrative consistency and demographic diversity