to-be/annomi-motivational-interviewing-therapy-conversations

--- license: openrail task_categories: - conversational language: - en tags: - medical - therapy - conversations - counseling - dialogues pretty_name: annomi size_categories: - n<1K --- # Dataset Card for Dataset Name <!-- Provide a quick summary of the dataset. --> Converted the AnnoMI motivational interviewing dataset into sharegpt format. It is the first public collection of expert-annotated MI transcripts. [Source](https://github.com/uccollab/AnnoMI). ## Dataset Details ### Dataset Description <!-- Provide a longer summary of what this dataset is. --> AnnoMI, containing 133 faithfully transcribed and expert-annotated demonstrations of high- and low-quality motivational interviewing (MI), an effective therapy strategy that evokes client motivation for positive change. Sample conversation number 40, regarding drug use: ``` client Pretty good. It's been three months now. therapist Nice. client Um, and— therapist Since the-- since you used, uh-- since the last time you used meth, right? client Right. therapist Okay. client Right. It's been three months, and I've been doing pretty well. therapist Good. client Um, and I met a couple new friends. therapist Nice. client So— therapist So-so, uh, have-- last time we talked, you know, you were kind of cutting the-the ex-boyfriend out. This is because he was your dealer, so-- client Right. therapist And how is that going for you? client Right. It's been good. I haven't answered the phone when he's called- therapist Okay. client -and a couple of my other friends have tried to call me as well, and I haven't answered the phone. Um, so it's been good for me to just reach out and like, meet a couple of new people, um, in the neigh-neighborhood that I'm living in. therapist So it sounds like you've made some good changes. The change in your environment, and some of your behaviors, and, uh, and these new friends. client Yeah, the new friends, and I've been using the, you know, the breathing exercises you taught me, when I get-- start to feel a little stressed out, and I've been- I've been using those and it helps me. I go for a walk and do my breathing exercises, and it calms me down and kind of gets me through that time. therapist So it sounds like-- ``` ### Dataset Sources <!-- Provide the basic links for the dataset. --> - **Repository:** [https://github.com/uccollab/AnnoMI] - **Paper [optional]:** [https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9746035] ## Uses <!-- Address questions around how the dataset is intended to be used. --> Training on longer conversations and medical advice. ### Direct Use <!-- This section describes suitable use cases for the dataset. --> [More Information Needed] ### Out-of-Scope Use <!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. --> [More Information Needed] ## Dataset Structure <!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. --> Sharegpt format to allow for long conversations back and forth. therapist is gpt client is human id is a concatenation of: - transcript_id: the unique numerical identifier of the conversation/transcript where this utterance belongs. Note that this identifier is NOT used for ordering, and it is only to distinguish between different conversations in the dataset. - video_url: the URL of the original video of the conversation/transcript where this utterance belongs. - topic: the topic(s) of the conversation/transcript where this utterance belongs. ``` [ { "id": "40_https://www.youtube.com/watch?v=wcu8oqHIsbc_reducing drug use", "conversations": [ { "from": "human", "value": "Pretty good. It's been three months now" }, { "from": "gpt", "value": "Nice." }, { "from": "human", "value": "Right. It's been three months, and I've been doing pretty well." }, { "from": "gpt", "value": "So-so, uh, have-- last time we talked, you know, you were kind of cutting the-the ex-boyfriend out. This is because he was your dealer, so--" }, ... ] ``` ## Dataset Creation ### Curation Rationale <!-- Motivation for the creation of this dataset. --> Despite the remarkable growth of research in recent years on the analysis of counselling conversations through natural language processing methods, the potential of this field has been greatly limited by the lack of access to publicly available therapy dialogues, especially those with expert annotations. #### Personal and Sensitive Information <!-- State whether the dataset contains data that might be considered personal, sensitive, or private (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.). If efforts were made to anonymize the data, describe the anonymization process. --> It is anonymized. ## Bias, Risks, and Limitations <!-- This section is meant to convey both technical and sociotechnical limitations. --> [More Information Needed] ### Recommendations <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. --> Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations. ## Citation [optional] <!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. --> @INPROCEEDINGS{9746035, author={Wu, Zixiu and Balloccu, Simone and Kumar, Vivek and Helaoui, Rim and Reiter, Ehud and Reforgiato Recupero, Diego and Riboni, Daniele}, booktitle={ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, title={Anno-MI: A Dataset of Expert-Annotated Counselling Dialogues}, year={2022}, volume={}, number={}, pages={6177-6181}, doi={10.1109/ICASSP43922.2022.9746035}} } ## Dataset Card Contact Toon Beerten ([e-mail])

原始信息汇总

数据集概述

• 名称: AnnoMI motivational interviewing dataset
• 格式: sharegpt格式
• 特点: 第一个公开的专家注释的MI（Motivational Interviewing）转录集合
• 来源: 未提供具体来源链接，仅提及为“Source”

搜集汇总

数据集介绍

构建方式

在心理治疗与自然语言处理的交叉领域中，高质量的对话数据集极为稀缺。该数据集基于AnnoMI项目，将133段经过专家精心标注的动机性访谈（Motivational Interviewing）会话转录，转换为ShareGPT格式。每一段对话都忠实保留了原始咨询场景中的治疗师与来访者互动，其中治疗师角色被设定为GPT、来访者为人类，并通过唯一标识符（transcript_id、视频URL和话题标签）对每条对话进行编码，以确保可追溯性和结构清晰。

特点

该数据集作为首个公开的专家标注动机性访谈语料库，具有鲜明的专业性和结构优势。它涵盖高、低质量的咨询示范，涉及药物滥用等敏感治疗话题，为分析咨询对话中的有效策略提供了稀缺资源。数据采用长对话格式，支持多轮次交互建模，且所有内容经过匿名化处理，兼顾了研究深度与隐私保护。

使用方法

研究者可直接加载该数据集，用于训练和评估心理咨询对话系统。数据遵循ShareGPT格式，便于与主流对话模型框架集成。使用时需注意其限定于英语语境，并聚焦于动机性访谈这一特定治疗流派。建议在模型微调或效果评估时，结合原始论文中的专家标注标准，以充分发挥其在医疗对话分析中的潜力。

背景与挑战

背景概述

在心理治疗与咨询领域，动机性访谈（Motivational Interviewing, MI）作为一种以患者为中心、旨在激发内在改变动力的循证疗法，已在药物滥用、慢性病管理等场景中展现出显著疗效。然而，长期以来，该领域的研究受限于高质量、专家标注的对话语料匮乏，阻碍了自然语言处理技术对治疗性对话的深度解析与自动化评估。2022年，由Wu Zixiu、Simone Balloccu等学者联合多所机构发布的AnnoMI数据集，首次公开了133段经专家标注的高低质量MI对话转录文本，涵盖药物使用、行为改变等主题。该数据集基于真实治疗视频转录，并经由两名独立专家依据MI治疗 fidelity 编码体系进行质量评估，为计算语言学与心理治疗的交叉研究提供了关键基准资源，其影响力已延伸至对话系统评估、治疗策略自动分析等前沿方向。

当前挑战

当前AnnoMI数据集面临的挑战主要源于两个层面。其一，在领域问题层面，心理治疗对话分析需应对高度情境化的语言现象，如隐喻、犹豫、情绪暗示等非结构化表达，现有自然语言处理模型难以精准捕捉治疗师与来访者之间的动态协作关系，尤其是对MI核心技巧（如开放式提问、肯定性反馈）的自动化识别仍存在显著偏差。其二，在数据集构建层面，样本规模仅133段对话，且主题集中于药物滥用与行为改变，导致模型泛化能力受限；此外，对话转录虽经匿名化处理，但原始视频中的非语言线索（如语调、肢体语言）的缺失，以及标注者间对治疗质量判断的主观差异，进一步增加了数据噪声与标注不一致性。这些限制要求后续研究需在扩充多领域、多模态语料的同时，探索更稳健的标注协议与跨文化适应性策略。

常用场景

经典使用场景

在心理治疗与计算语言学交叉领域，AnnoMI数据集作为首个公开的专家标注动机访谈语料库，最经典的使用场景是训练和评估面向治疗对话的自然语言处理模型。研究者利用该数据集中133段高质量与低质量的动机访谈转录文本，构建能够自动识别治疗师咨询技巧（如开放式提问、反映性倾听）以及评估对话治疗质量的分类与回归模型，从而推动计算精神病学中对话智能体的发展。

解决学术问题

该数据集解决了长期困扰学术界的核心难题——缺乏公开可获取且经过专家标注的治疗对话数据。此前，心理咨询领域的语料库多因隐私限制而无法共享，导致基于深度学习的治疗对话分析研究进展缓慢。AnnoMI的出现使得研究者能够系统性地探索治疗对话的语言特征与治疗效果之间的关联，为开发自动化的治疗质量评估工具和循证心理干预方案提供了坚实的实证基础，显著推动了计算语言学在心理健康领域的应用研究。

衍生相关工作

围绕AnnoMI数据集，学术界涌现了一系列经典衍生工作，包括基于预训练语言模型（如BERT、RoBERTa）的治疗对话质量分类器、融合情感计算与对话行为识别的多任务学习框架，以及利用对比学习增强治疗技巧表征的模型。同时，该数据集也启发了跨语种治疗对话数据集的构建方法，并催生了面向治疗师共情能力自动评估的基准测试任务，形成了从数据构建到模型评估的完整研究链条。