时间：2023年12月7日（星期四）15:00-16:00
地点：经管大楼A楼 四楼第二会议室报告厅
主题：整合物品效用与用户兴趣的可信推荐研究（TrustworthyRecommendation Research By Integrating Item Utility and User Interest）
主讲人：尹裴（beat365中国官方网站）
简介：尹裴，女，晨光学者，副教授，硕导，研究方向：智能决策、人机协同，主持及参与国家自然科学基金、教育部人文社科项目、上海市晨光计划、上海市哲学社科项目等国家级、省部级项目10余项，发表学术论文50余篇，其中SCI/SSCI/EI检索30余篇，专著1部，教材3本。
Dr. Pei Yin , female, a ‘Chen Guang’ Scholar and an Associate Professor, also serves as a master's supervisor. Her research focuses on intelligent decision-making and human-machine collaboration. She has led and participated in more than ten national and provincial-level projects, including those funded by the National Natural Science Foundation, the Ministry of Education's Humanities and Social Science Project, the Shanghai ‘Chen Guang’ Plan, and the Shanghai Philosophy and Social Science Project. She has published over 50 academic papers, with more than 30 indexed in SCI/SSCI/EI, one monograph, and three textbooks.
摘要：现有推荐模型主要侧重于为用户推荐其感兴趣的物品，却忽略了评估物品对用户的实际效用，从而可能导致如“诱导沉迷”等算法偏见问题。此外，现有算法过分关注用户历史行为中其频繁接触且高度相似的兴趣点，这可能使用户陷入“信息茧房”。因此，本研究旨在解决如何向用户推荐既符合其兴趣又实际有用的物品这一核心问题，并从整合物品效用和用户兴趣的角度出发，综合运用多任务学习、表征学习和生成对抗网络等多种方法，深入挖掘影响用户决策的多种因素。该研究已在健康医疗产品和众包任务等不同的推荐场景中得到应用，并通过一系列实验验证了推荐结果的可信度，包括其精准度、多样性和有用性等方面。
The extent recommendation models primarily focus on recommending items that users are interested in, yet they often overlook the utility of these items for users. This oversight can lead to issues like algorithmic biases, such as inducing addiction. Furthermore, these algorithms tend to overemphasize interest points that are frequently encountered and highly similar in users' historical behaviors, potentially causing Filter Bubble.
Therefore, this study aims to address the critical issue of how to recommend items that are both of interest and genuinely useful to users, and thus proposes an approach that integrates item utility and user interests. By employing a combination of methods such as multi-task learning, representation learning, and generative adversarial networks, this research delves deeply into various factors that influence user decision-making. The study has been applied in diverse recommendation scenarios, including healthcare products and crowdsourcing tasks. It has been validated through a series of experiments, confirming the credibility of the recommendations in terms of accuracy, diversity, and usefulness.
