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

职业：走向对话式推荐系统：可解释性、公平性和人在环学习

• 批准号: 2046457
• 负责人: Yongfeng Zhang
• 金额: $ 54.97万
• 依托单位: Rutgers University New Brunswick
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-10-01 至 2026-09-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2046457&HistoricalAwards=false
• 关键词: CAREER; Towards; Conversational; Recommendation; Systems

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.

人工智能 (AI) 的最新进展积累了丰富的信息检索、自然语言处理和个性化推荐模型工具箱。通过对基准数据集进行优化，许多模型的开发都考虑了算法，而不是把人类作为中心考虑因素。但人工智能的最终目的是服务人类、与人类合作、最终造福人类。因此，人工智能的算法方法必须让人类参与模型设计、实施和验证的循环。该项目专注于对话式人工智能，这是一种让人类参与循环的有前途的方法，它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来帮助用户进行信息搜索和决策。它将开发可解释且具有公平意识的对话推荐算法。作品和演示的展示将有助于吸引对计算研究感兴趣的更广泛的受众。通过将透明度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中，该项目的结果将教育学生了解人工智能如何不仅有用，而且具有社会责任感。该项目将开发一个对话推荐的通用框架，连接自然语言理解和对话状态管理。在此框架下，项目将探索三个方向。第一个方向旨在开发基于人机协作推理的可解释对话策略，为用户带来认知便利，并有助于建立人与人工智能之间的信任。第二个方向探索基于短期和长期公平学习的公平感知对话策略，这有助于实现优势用户和弱势用户之间的公平推荐体验。第三个方向旨在开发一种会话推荐的学习评估协议，它结合了在线众包和离线模型学习评估的优点。该项目还将开发一个原型对话推荐平台作为课堂项目，以支持负责任的人工智能教育。该项目将导致共享数据和评估平台向信息检索、数据挖掘、推荐系统和更广泛的人工智能社区的传播。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5)

• DOI: 10.1145/3523227.3546767
• 发表时间: 2022-03
• 期刊: Proceedings of the 16th ACM Conference on Recommender Systems
• 作者: Shijie Geng;Shuchang Liu;Zuohui Fu;Yingqiang Ge;Yongfeng Zhang

System 1 + System 2 = Better World: Neural-Symbolic Chain of Logic Reasoning

• DOI: 10.18653/v1/2022.findings-emnlp.42
• 发表时间: 2022
• 作者: Wenyue Hua;Yongfeng Zhang

User-Controllable Recommendation via Counterfactual Retrospective and Prospective Explanations

• DOI: 10.48550/arxiv.2308.00894
• 发表时间: 2023-08
• 作者: Juntao Tan;Yingqiang Ge;Yangchun Zhu;Yinglong Xia;Jiebo Luo;Jianchao Ji;Yongfeng Zhang

AutoLossGen: Automatic Loss Function Generation for Recommender Systems

• DOI: 10.1145/3477495.3531941
• 发表时间: 2022-04
• 期刊: Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval
• 作者: Zelong Li;Jianchao Ji;Yingqiang Ge;Yongfeng Zhang

Graph Collaborative Reasoning

• DOI: 10.1145/3488560.3498410
• 发表时间: 2021-12
• 期刊: Proceedings of the Fifteenth ACM International Conference on Web Search and Data Mining
• 作者: H. Chen;Yunqi Li;Shaoyun Shi;Shuchang Liu;He Zhu;Yongfeng Zhang