大数据环境下融合多源信息的推荐系统关键问题研究

In the big data environment, the variety and heterogeneity of information has becoming a great challenge for recommender systems. Rising to this challenge, we propose in this project to study critical issues of multi-source information fusion for recommender systems. We study such critical issues as network construction, network integration for multi-source information and information-fusion based recommendation, and then propose effective solutions to solve them. From the viewpoint of network analysis, we propose to convert multiple information sources into multiple networks. Then, we propose two strategies as network fusion and information fusion to address such critical issuses. First, we suggest to integrate the multiple networks and then develop new recommendation methods from the perspective of statistics and graph theory. Second, we propose to directly fuse these information sources and establish statistics and machine learning methods for recommendation. Finally, we use the proposed methods to certain application fields and select several solutions with better performances. We study the crucial issues of how to fuse multi-source information for recommendation based on such opinions as systems, integrity and quantification. The objective of this research is to more sufficiently utilize multiple information sources, for the purpose of improving recommendation accuracy, novelty and diversity. We places great importance on statistical methods and graph theory with a solid foundation, and thus will provide helpful experiences in future applications of these theories and methods to crucial scientific issues in information management field.

本项目针对大数据环境下推荐系统面临的信息多样化和异质性挑战，提出大数据环境下融合多源信息的推荐系统关键问题研究。项目研究推荐系统多源信息的网络构建、多源信息的网络融合以及基于信息融合的推荐等关键问题，并提出有效的解决途径。项目首先从网络分析的视角将多种信息转化为多个网络。然后从网络融合和信息融合两条思路进行研究。基于网络融合的思想，提出融合多个网络的方法，以及基于网络融合进行推荐的统计学和图论方法。基于信息融合的思想，建立直接基于多源信息进行推荐的统计学和机器学习方法。最后将所提出方法应用于特定应用领域，筛选效果较好的解决方案。项目从系统的、整合的、量化的观点研究大数据环境下融合多源信息的推荐系统关键问题，旨在充分利用多种信息提高推荐的精确性、新颖性和多样性。项目推广应用具有坚实理论基础的统计学、机器学习和图论方法，为运用这些方法解决信息管理领域的重要科学问题提供有益借鉴。

本项目围绕大数据环境下推荐系统信息融合分析中的若干关键科学问题，遵循理论与应用相结合的研究思路，从信息科学与系统科学的角度，应用统计分析、智能计算等国际领先的理论方法，考察推荐系统中多个来源不同性质的信息作为整体所遵循的规律，开展复杂网络关联推断问题的智能化方法等研究，构建融合多源信息的推荐系统异质网络随机游走、异质网络回归分析、异质网络融合等方法，攻克多种信息源缺失数据表示以及网络信息融合等难题。项目取得一系列创新性研究成果，在大数据环境下推荐系统领域具有重要的理论意义；同时，项目扩展应用成果在电子商务、社交网络、信息推荐等领域产生重要实际应用价值。成果方面，在Decision Support Systems, Future Generation Computer Systems, Information Systems Frontiers, BMC Systems Biology, PLOS ONE, World Wide Web Journal, BMC Systems Biology, Journal of Computer Science and Technology, Expert Systems with Applications等相关领域SCI检索的国外重要学术期刊发表论文11篇，在Hawaii International Conference on System Sciences (HICSS), Pacific Asia Conference on Information Systems (PACIS) 等国际会议发表英文论文5篇。