项目名称: 面向推荐系统中异构隐式反馈建模的迁移学习技术研究
项目编号: No.61502307
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 其他
项目作者: 潘微科
作者单位: 深圳大学
项目金额: 21万元
中文摘要: 智能推荐技术作为应对信息过载问题和提供个性化服务的有效手段,具有重要的研究价值和广泛的应用需求。然而,推荐算法的研究往往限于用户对物品的同构显式反馈(如评分)和同构隐式反馈(如购买),较少涉及异构隐式反馈等更为常见的问题。异构隐式反馈(HIF)建模作为推荐系统领域的一个重要发展方向,旨在挖掘多种隐式反馈(如购买、浏览等)背后的用户偏好,进而提供更好的个性化服务。目前,HIF的研究成果还很少,面临着用户真实偏好的不确定性、用户反馈的异构性和稀疏性等挑战。为此,本项目拟研究相应的迁移学习技术来深入挖掘用户的真实偏好,为HIF中的关键问题提供新的数据建模方法和算法设计思路。本项目拟重点研究面向HIF的“迁移学习模式”、“混合迁移学习”和“多维度效果评估”等几个方面,所提出的理论和算法,可望丰富迁移学习、偏好处理、推荐系统等领域的研究成果,并为HIF在电子商务等互联网推荐系统中的应用提供解决方案。
中文关键词: 迁移学习;隐式反馈;异构反馈
英文摘要: Intelligent recommendation techniques have been an effective way to address the information overload challenge and to provide personalization services. However, most recommendation algorithms focus on users' homogeneous explicit feedbacks such as ratings and homogeneous implicit feedbacks such as transactions instead of heterogeneous implicit feedbacks. As an important research direction, modeling of heterogeneous implicit feedbacks (HIF) has the potential to mine users' preferences from users' different types of implicit feedbacks such as transactions and examinations, so as to provide better personalization services. So far, there have been very few research works on HIF, which is associated with some fundamental challenges, e.g., high uncertainty of users' true preferences, and heterogeneity and sparsity of users feedbacks. As a response, in this project, we aim to study the HIF problem and design some transfer learning techniques to mine users' true preferences, which shall provide new data modeling methods and algorithm design ideas for HIF. Specifically, we aim to study three problems for HIF, including knowledge transfer styles, mixed transfer learning, and multidimensional evaluation. The proposed theory and algorithms are expected to enrich the research works on transfer learning, preference handling and recommender systems, and to provide a solution for real deployment of HIF in e-commerce and other internet recommender systems.
英文关键词: Transfer Learning;Implicit Feedbacks;Heterogeneous Feedbacks