Researchers propose behaviorally grounded user profiles from social media to personalize language models

Researchers introduce profile behavioral grounding, a framework that extracts open-ended, high-fidelity user profiles directly from authentic, anonymized social media posts. The approach aims to address limitations in existing persona-driven techniques, which often flatten individual variation and rely on stereotypes rather than nuanced signals of human preferences.

The framework evaluates profiles across two paradigms: train-time personalization via supervised finetuning (SFT) and non-parametric test-time multi-perspective reasoning. In experiments on complex recommendation and open-ended query benchmarks, behaviorally grounded profiles consistently improved base models and outperformed synthetic profile baselines.

The authors report that behavior-derived profiles drive stronger parametric alignment and enable richer, multifaceted reasoning compared to synthetic alternatives. They argue that open-ended, behavior-based profiles provide a more diverse and effective foundation for next-generation personalized language systems.

The work is documented in an arXiv preprint and the associated codebase has been released under an open-source license for further research and development.