RongxinChen/MBTI_dpo_e_i

Multi-Personality Generation (MPG) Datasets

**Paper** | **GitHub**

This repository contains datasets released as part of the paper "Multi-Personality Generation of LLMs at Decoding-time", which was accepted at WSDM 2026.

Introduction

The Multi-Personality Generation (MPG) framework enables Large Language Models to simultaneously embody multiple personalization attributes during decoding without requiring extra training. It leverages implicit density ratios in single-dimensional models and implements Speculative Chunk-level based Rejection sampling (SCR) to significantly reduce computational overhead while maintaining high-quality generation.

Dataset Description

The release includes Direct Preference Optimization (DPO) datasets used for MBTI personality simulation and Role-Playing scenarios.

MBTI DPO Datasets

These datasets focus on specific Myers-Briggs Type Indicator dimensions:

• —E / I: Extraversion vs. Introversion
• —J / P: Judging vs. Perceiving
• —T / F: Thinking vs. Feeling
• —S / N: Sensing vs. Intuition

RolePlay DPO Datasets

• —Personality: DPO data for general personality alignment.
• —Profile: DPO data based on character profiles.

Citation

If you find our work or datasets useful, please consider citing our paper:

@inproceedings{chen2026multipersonality,
  title={Multi-Personality Generation of LLMs at Decoding-time},
  author={Chen, Rongxin and Li, Yunfan and Yuan, Yige and Xu, Bingbing and Shen, Huawei},
  booktitle={Proceedings of the 19th ACM International Conference on Web Search and Data Mining (WSDM '26)},
  year={2026},
  publisher={ACM},
  url={http://arxiv.org/abs/2511.01891}
}