CAREER: Towards Conversational Recommendation Systems: Explainability, Fairness, and Human-in-the-Loop Learning

人工智能（AI）的最新进展为信息检索、自然语言处理和个性化推荐积累了丰富的模型工具箱。通过对基准数据集进行优化，许多模型都是在考虑算法的情况下开发的，而不是把人作为中心考虑因素。然而，人工智能的最终目标是为人类服务，与人类合作，最终造福人类。因此，人工智能的算法方法必须将人类置于模型设计、实施和验证的循环中。该项目侧重于对话式人工智能，这是一种将人类置于循环中的有前途的方法，可以实现人类和人工智能之间的直接对话，以进行模型学习。特别是，该项目探索会话推荐系统，以帮助用户寻求信息和决策。它将为会话推荐开发可解释和公平感知的算法。介绍工作和演示将有助于吸引对计算研究感兴趣的更广泛的受众。通过将透明和公平原则融入信息检索、数据挖掘和人工智能等领域的计算机科学课程，该项目的成果将教育学生了解人工智能如何不仅有用，而且对社会负责。该项目将为会话推荐开发一个通用框架，将自然语言理解和对话状态管理连接起来。在此框架下，项目将探索三个方向。第一个方向是基于人机协同推理开发可解释的对话策略，为用户带来认知上的便利，并有助于建立人与AI之间的信任。第二个方向探索基于短期和长期公平学习的公平感知会话策略，有助于实现优势用户和劣势用户之间的公平推荐体验。第三个方向旨在开发一种用于会话推荐的学习评估协议，该协议结合了在线众包和离线模型学习的优势进行评估。该项目还将开发一个原型对话推荐平台，作为一个班级项目，以支持负责任的人工智能教育。该项目将为信息检索、数据挖掘、推荐系统和更广泛的人工智能社区提供共享数据和评估平台。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Recent advances in Artificial Intelligence (AI) have accumulated a rich toolbox of models for information retrieval, natural language processing and personalized recommendation. By optimizing over benchmark datasets, many of the models were developed with an algorithmic consideration instead of putting human as the central consideration. However, the ultimate goal of AI is to serve humans, collaborate with humans, and, ultimately, benefit humans. As a result, algorithmic approaches to AI must put humans in the loop for model design, implementation and validation. This project focuses on conversational AI, a promising approach towards putting humans in the loop, which enables direct conversation between human and AI for model learning. In particular, the project explores conversational recommender systems to help users in information seeking and decision making. It will develop explainable and fairness-aware algorithms for conversational recommendation. Presentation of the work and demos will help to engage with wider audiences that are interested in computational research. By integrating transparency and fairness principles into computer science courses on areas such as Information Retrieval, Data Mining and Artificial Intelligence, results from the project will educate students to understand how AI can be not only useful but also socially responsible.This project will develop a general framework for conversational recommendation that bridges natural language understanding and dialog state management. With the framework, the project will explore three directions. The first direction aims at developing explainable conversation strategies based on human-machine collaborative reasoning, which brings cognitive ease to users and helps to build trust between human and AI. The second direction explores fairness-aware conversation strategies based on short-term and long-term fairness learning, which helps to achieve fair recommendation experiences between advantages and disadvantaged users. The third direction aims at developing a learning to evaluate protocol for conversational recommendation, which unifies the advantages of online crowd-sourcing and offline model learning for evaluation. The project will also develop a prototype conversational recommendation platform as a class project to support the education of responsible AI. The project will result in the dissemination of shared data and evaluation platforms to the Information Retrieval, Data Mining, Recommender System, and broader AI communities.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.