MPCCD-MLF: Multi-Round Professional Consultation Dialogue Dataset for Medical, Legal and Financial Domains

The MPCCD-MLF multi round professional consultation dialogue dataset consists of 32175 high-quality multi round dialogue samples, covering three professional fields: healthcare, law, and finance. Each sample is derived from a real professional consulting scenario and adopts a unified multi round interaction structure, which includes at least continuous rounds such as "user questioning expert answering user follow-up expert re answering", used to depict the process of gradually clarifying questions and making decisions in professional consulting. The dataset is stored in standard JSON format, with each sample containing two core fields: system and conversations. Among them, the system field is used to define the domain to which the conversation belongs, expert roles, and response standards; The conversations field is a list of multiple rounds of interactions arranged in chronological order, with each round containing two sub fields: from (message source, valued as User or Expert) and value (corresponding to the conversation text content). All samples have undergone cleaning, desensitization, explicit reconstruction of the thought chain, and quality review of the combination of models and manual methods to ensure high consistency in logical coherence, professionalism, and security of the data. They are suitable for instruction fine-tuning and multi round inference training of large language models in professional fields.

搜集汇总

背景与挑战

背景概述

MPCCD-MLF是一个包含32175个高质量多轮对话样本的专业咨询对话数据集，覆盖医疗、法律和金融三个领域。每个样本采用统一的多轮交互结构（用户提问-专家回答-用户追问-专家再回答），并经过清洗、脱敏和思维链显式重建等严格质量审核，确保逻辑连贯性和专业性，适用于大语言模型在专业领域的指令微调与多轮推理训练。