库帕思高质量政务服务思维链（Chain-of-Thought）数据集

本数据集面向需要复杂推理能力的政务服务大模型及AI助手研发团队，提供高质量思维链训练语料。基于真实政务场景中经人工核验的2500组标准问答对，通过大模型对市民提问进行多步骤推理生成，完整还原“问题确认→诉求分析→政策查询→回复策略”的政务处理逻辑链条，并已通过专业人员核对。每条样本均包含市民原始提问、结构化思维链推理过程和生成的口语化回复，覆盖政策咨询、投诉处理等多元场景。数据集已按3:1比例划分为训练集与验证集，可直接用于思维链微调与推理能力强化。

This dataset is tailored for R&D teams developing government service large language models (LLMs) and AI assistants that require complex reasoning capabilities, providing high-quality chain-of-thought (CoT) training corpora. It is generated via multi-step reasoning conducted by LLMs on citizens' inquiries, based on 2,500 standard question-answer pairs manually verified in real government service scenarios, fully restoring the logical workflow of government affairs processing: "problem confirmation → demand analysis → policy query → response strategy", and has been validated by professional personnel. Each sample includes the citizen's original inquiry, structured chain-of-thought reasoning process, and generated colloquial response, covering diverse scenarios such as policy consultation and complaint handling. The dataset has been split into training and validation sets at a 3:1 ratio, and can be directly applied to chain-of-thought fine-tuning and reasoning capability enhancement.

搜集汇总

背景与挑战

背景概述

该数据集专为政务服务大模型及AI助手研发设计，提供基于2500组真实政务场景标准问答对的高质量思维链训练语料，覆盖政策咨询、投诉处理等多元场景。数据集已划分为训练集与验证集，可直接用于思维链微调与推理能力强化。