基于多示例学习的多模态信息表达与推荐方法研究

项目名称： 基于多示例学习的多模态信息表达与推荐方法研究

项目编号： No.71201120

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 管理科学与工程

项目作者： 袁汉宁

作者单位： 武汉理工大学

项目金额： 22万元

中文摘要： 互联网中存在多种模态的媒体对象，单一模态的信息表达限制了用户获取信息的有效性和准确性。多示例学习适合解决多模态信息的表达问题，但其现有的方法不能处理多模态信息。本项目研究基于多示例学习的多模态信息表达和推荐方法： (1)研究多模态信息的多示例表达机制，构建异质多模态信息表达模型和同质多模态信息的表达模型，以全面准确地表达多模态信息； (2)研究多模态信息的多示例学习算法，设计基于示例选择的启发式异质多示例SVM分类算法，以及基于多模态信息统一化方法的同质多示例学习算法，实现多模态信息的比较和计算； (3)研究多模态信息的多示例推荐算法，实现包层次粗粒度的推荐和示例层次细粒度推荐；设计用户兴趣更新算法及时跟踪用户兴趣，用云模型可视化用户兴趣。 本项目最后研制原型系统验证方法的有效性和可靠性，成果将为用户全面准确地获取信息资源提供理论和方法支持，具有潜在的应用价值。

中文关键词： 多示例学习；多模态信息；推荐；信息表达；

英文摘要： There are multi-modalities of media objects in Internet. The representation of single modality information limits the validity and accuracy of information user accesses. Multi-instance learning(MIL) is suitable for solving the problems to represent multi-modality information,but its existing algorithms cannot deal with multi-modality information. In this project, the methodology of multi-modality information representation and recommendation will be studied on the basis of MIL. (1)Based on MIL, the representation mechanism of multi-modality information will be researched to build up heterogeneous multi-modality information representation model and homogeneous multi-modality information representation model, for representing multi-modality information comprehensively and accutately. (2)The MIL algorithms for multi-modality information will be designed. Heuristic heterogeneous SVM classification algotrithm based on instance selection and homogeneous MIL algotrithm based on uniform method, are presented to compare and compute the multi-modality information. (3)The recommendation algorithms of multi-modality information will be designed to recommend on both bag level with rough granularity and instance level with fine granularity. The algorithms of updating user interest will be studied to track user interest

英文关键词： multi-instance learning；multi-modality information；recommendation；information expression；