CRII: HCC: Empowering Human-AI Collaboration through Conversational Explanations

该奖项的全部或部分资金来自《2021年美国救援计划法案》(公法117-2)。研究表明，提供有关推荐系统的解释会对用户体验产生积极影响。因此，开发人员在社交媒体等应用程序中采用了许多可解释的推荐模型和界面。然而，这些解释对于具有不同数字素养或计算知识的用户来说并不是个性化的，可能并不总是确保用户理解有助于数据或算法的基本原理。这个项目通过探索用户对基于人工智能的、可解释的推荐系统的理解的动态过程以及这种理解是如何演变的，来填补这一研究空白。此外，通过研究一种新颖的人与目标系统之间的会话交互，扩展了现有的可解释推荐系统的研究视野。最后，这项工作将通过一个统一的研究计划来实现一个新的生态系统，该研究计划包括为不同的用户建模解释策略和探索和开发实用的系统原型。该项目的成功将加深对为不同背景和专业知识的用户设计和实施公平和透明的日常使用人工智能系统的科学理解。例如，用户可能会从健康推荐系统中寻求个性化的解释，并在透明度和理解力的基础上做出明智的决定。该项目旨在设计和开发一个对话代理，以提供基于人工智能的结果和建议的个性化解释。这项工作将使用参与式设计来探索与机器生成的解释相关的用户心理模型，开发一个对话代理来得出用户的心理模型，并基于推断的模型提供个性化的解释。本项目旨在实现基于最先进的可解释混合式推荐系统的三个目标。首先，该项目进行了一个以用户为中心的参与式设计焦点小组，以测量和评估用户与文本、主题和社交推荐模型的多层次粒度解释交互的心理模型。其次，这个项目利用以人为中心的设计方法来开发一个原型，它允许动态启发机制并提供对话解释。第三，该项目采用混合方法调查和理论化用户在混合推荐系统中与开发的基于对话的解释交互的体验和理解。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).Research has demonstrated that providing explanations about recommendation systems positively affect users' experiences. Accordingly, developers have adopted many explainable recommendation models and interfaces in applications such as social media. However, these explanations are not personalized to users with varied digital literacy or computational knowledge and may not always ensure users understand the underlying rationale contributing to data or algorithms. This project fills this research gap by exploring the dynamic process of a user's understanding of an AI-based, explainable recommender system and how this understanding evolves. In addition, it extends the existing research horizon on explainable recommender systems by investigating a novel conversational interaction between people and target systems. Finally, this work will enable a new ecosystem through a unified research program consisting of modeling explanation strategies for different users and exploring and developing practical system prototypes. The project's success will deepen the scientific understanding of designing and implementing fair and transparent everyday use AIsystems for users of varied backgrounds and expertise. For instance, users may seek personalized explanations from a health recommendation system and make an informed decision based on transparency and comprehension.This project aims to design and develop a conversational agent to provide personalized explanations around AI-based results and recommendations. The work will use participatory design to explore user mental models regarding machine-generated explanations, develop a conversational agent to elicit the user's mental model, and provide personalized explanations based on the inferred model. This project aims to achieve three objectives based on a state-of-the-art explainable hybrid recommender system. First, this project conducts a user-centric participatory design focus group to measure and estimate the user's mental model of interacting with multi-level granularity explanations for text, topic, and social recommendation models. Second, this project utilizes a human-centered design approach to develop a prototype that allowsdynamical elicitation mechanisms and provides conversational explanations. Third, this project adopts mixed methods to investigate and theorize the user's experience andcomprehension of interacting with the developed conversation-based explanation in hybrid recommender systems.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.