Persuading to Prepare for Quitting Smoking with a Virtual Coach: Using States and User Characteristics to Predict Behavior - Data, Analysis Code and Appendix

This repository contains the data, analysis code, and appendix of the paper "Persuading to Prepare for Quitting Smoking with a Virtual Coach: Using States and User Characteristics to Predict Behavior" by Nele Albers, Mark A. Neerincx, and Willem-Paul Brinkman, published in <em>Proceedings of the 22nd International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2023)</em>.<br><strong>Data</strong>The paper is based on data collected during a study on the online crowdsourcing platform Prolific run between 20 May 2021 and 30 June 2021. The Human Research Ethics Committee of Delft University of Technology granted ethical approval for the research (Letter of Approval number: 1523).In this study, smokers who were contemplating or preparing to quit smoking interacted with the text-based virtual coach Sam in up to five conversational sessions. In each session, participants were assigned a new preparatory activity for quitting smoking, such as thinking of and writing down reasons for quitting smoking. Since becoming more physically active may make it easier to quit smoking, half of the activities addressed becoming more physically active. The virtual coach chose from five persuasive strategies to persuade people to do their activity. In the first two sessions, the persuasive strategy was chosen uniformly at random; in the last three sessions, the persuasive strategy was determined by a persuasion algorithm that differed between four conditions. In the next session, participants were asked to indicate the effort they spent on their activity, which served as basis for the reward signal for the persuasion algorithm. The study was pre-registered in the Open Science Framework (OSF): https://osf.io/k2uac. This pre-registration describes the study design, measures, etc. Note that the data we provide here is only a part of the data collected in the study, namely, the data related to studying the prediction of behavior (i.e., the effort people spent on their activities) based on user states and characteristics.<br><strong>Analysis Code</strong>Our analysis can be reproduced using Docker and Jupyter Notebook. We provide instructions for this in the README-files accompanying our analysis code.<br><strong>Appendix</strong>We also provide the Appendix of our paper, which contains more information on the virtual coach (including the conversation structure and preparatory activities), persuasion algorithm, data collection, optimal and worst policies computed for research questions Q3 and Q4, and the weighting of samples based on similarity for research question Q6.Regarding the preparatory activities, note that there were two different formulations: one for during the session, and one for the reminder message people received on Prolific.The former asked people to do the activity "<em>after</em> this session" and told people that they would receive the video link in the Prolific reminder message in case the activity involved watching a video; the latter asked people to do the activity "before the next session" in sessions 1-4 and contained the video link in case the activity involved watching a video. All activity formulations can be found together with the virtual coach code: https://github.com/PerfectFit-project/virtual_coach_rl_persuasion_algorithm/blob/main/Activities.csv. Custom action code further modifies the reminder message activity formulation for session 5, which is the last session (https://github.com/PerfectFit-project/virtual_coach_rl_persuasion_algorithm/blob/main/actions/actions.py).<br><strong>Further Resources</strong>Here are some pointers to further resources:This is the link to the paper: https://www.ifaamas.org/Proceedings/aamas2023/pdfs/p717.pdf.Data on the acceptance of the virtual coach from the same study can be found here: https://doi.org/10.4121/19934783.v1.Data on users' needs for a digital smoking cessation application from the same study can be found here: https://doi.org/10.4121/20284131.v2.Data on users' action plans for doing the activities (n = 469) and free-text responses to reflective questions about the activities (n = 2026) from the same study is available here: https://doi.org/10.4121/21905271.v1.The implementation of the virtual coach Sam is available here: https://doi.org/10.5281/zenodo.6319356. Journal paper describing the persuasion algorithm and analyzing its effectiveness: https://doi.org/10.1371/journal.pone.0277295.If you have questions about the data, analysis code, or appendix, please contact Nele Albers ([e-mail]).<br>

本仓库包含论文"Persuading to Prepare for Quitting Smoking with a Virtual Coach: Using States and User Characteristics to Predict Behavior"的相关数据、分析代码与附录，该论文由Nele Albers、Mark A. Neerincx与Willem-Paul Brinkman合著，发表于<em>第22届自主代理与多代理系统国际会议（AAMAS 2023）论文集</em>。<br><strong>数据集</strong>本论文基于2021年5月20日至2021年6月30日期间在Prolific众包平台开展的一项研究收集的数据。代尔夫特理工大学人类研究伦理委员会为该研究授予伦理许可（批准函编号：1523）。本研究中，有戒烟意向或正筹备戒烟的吸烟者与基于文本的虚拟教练（Virtual Coach）Sam进行至多五轮对话交互。每一轮会话中，参与者都会获得一项全新的戒烟筹备任务，例如思考并写下戒烟理由。鉴于提升身体活动量或可辅助戒烟，半数任务围绕提升身体活动量设计。虚拟教练将从五种说服策略中选取一种，引导参与者完成任务。在前两轮会话中，说服策略采用均匀随机选取方式；在后三轮会话中，说服策略由四种不同条件下的说服算法确定。在下一轮会话开始前，参与者需报告其完成任务所投入的精力，该数据将作为说服算法奖励信号的计算基础。本研究已在开放科学框架（Open Science Framework, OSF）完成预注册：https://osf.io/k2uac。该预注册文件详细阐述了研究设计、测量方法等内容。需注意，本仓库提供的数据仅为该研究收集数据的子集，即与基于用户状态与特征预测行为（即参与者完成任务所投入的精力）相关的数据。<br><strong>分析代码</strong>本研究的分析流程可通过Docker与Jupyter Notebook复现，相关操作说明已收录于分析代码附带的README文件中。<br><strong>附录</strong>本仓库同时提供论文附录，其中包含更多关于虚拟教练（含对话结构与筹备任务）、说服算法、数据收集流程、针对研究问题Q3与Q4计算得到的最优与最差策略，以及针对研究问题Q6基于相似度的样本加权方法的相关信息。关于筹备任务，需注意存在两种不同的表述形式：一种用于会话过程中，另一种用于参与者在Prolific平台接收的提醒消息内。前者要求参与者<em>在本次会话结束后</em>完成任务，并说明若任务涉及观看视频，将在Prolific平台的提醒消息中提供视频链接；后者要求参与者在第1至4轮会话中<em>在下一次会话开始前</em>完成任务，且若任务涉及观看视频，则会在消息中附带视频链接。所有任务表述均可在虚拟教练代码仓库中查阅：https://github.com/PerfectFit-project/virtual_coach_rl_persuasion_algorithm/blob/main/Activities.csv。自定义动作代码还针对最后一轮会话（即第5轮）的提醒消息任务表述进行了修改，相关代码可访问：https://github.com/PerfectFit-project/virtual_coach_rl_persuasion_algorithm/blob/main/actions/actions.py。<br><strong>补充资源</strong>以下为相关补充资源链接：本论文原文链接：https://www.ifaamas.org/Proceedings/aamas2023/pdfs/p717.pdf。本研究中关于虚拟教练接受度的数据可在此获取：https://doi.org/10.4121/19934783.v1。本研究中关于用户对数字化戒烟应用的需求数据可在此获取：https://doi.org/10.4121/20284131.v2。本研究中关于用户完成任务的行动计划（样本量n=469）以及针对任务的反思性问题自由文本回复（样本量n=2026）的数据可在此获取：https://doi.org/10.4121/21905271.v1。虚拟教练Sam的实现代码可在此获取：https://doi.org/10.5281/zenodo.6319356。一篇阐述说服算法并分析其有效性的期刊论文：https://doi.org/10.1371/journal.pone.0277295。若您对本仓库的数据、分析代码或附录存在疑问，请联系Nele Albers（邮箱：[e-mail]）。