面向云计算的高精度多粒度信誉评价机制研究

With growth of the cloud computing, the importance of the reputation system for cloud computing services has ever attracted more attentions. Establishing an objective and reliable reputation evaluation mechanism has been becoming a critical factor to promote cloud computing development. Reputation evaluation mechanism has been successfully applied in P2P, e-commerce and other fields, but the diversity of services, incompleteness of user recommendation, and complexity of malicious attack in cloud computing have put forward higher requirements to the accuracy and granularity of the reputation evaluation mechanism. This project intends to carry out research on high-precision multi-granularity reputation evaluation mechanism for cloud computing: concentrate on theoretical analysis of the reputation evaluation mechanism, which abstracts basic problems and formal representation of various current reputation evaluation mechanisms, and analyzes the cruxes which affect the accuracy of reputation calculation, guiding the design of the reputation evaluation mechanism; propose a rating prediction based index weight calculation method, which takes account into users rating distribution to fill incomplete user ratings, and derive the relationship between rating indices; present a high-precision multi-granularity reputation calculation model, which considering the relationship of multiple related ratings of referees, eliminates malicious and unreliable recommendation for accuracy improvement in the reputation calculation, and assesses cloud service reputation with multi-granularity according to the cloud application requirements. The research will provide a flexible and accurate reputation evaluation mechanism for applications such as service recommendation etc in cloud computing environment.

随着云计算业务的增长，人们开始意识到云计算环境下服务评价的重要性，建立一个客观可靠的信誉评价机制已经成为影响云计算推广的一个重要因素。信誉评价已经在P2P、电子商务等领域得到成功应用，但是云计算环境下服务的多样性、用户推荐的不完整性以及恶意攻击的复杂性等给信誉评价的精度和粒度提出了更高层次的要求，本课题拟开展面向云计算的高精度多粒度信誉评价机制研究：对信誉评价机制进行理论分析，抽象出信誉评价机制的基本问题和形式化表示，分析影响信誉值计算精度的关键因素，作为信誉评价机制设计的理论依据；提出一种基于推荐预测的指标权重计算方法，根据用户推荐的分布特征填充用户的不完整评价，推出评价指标之间的关系；提出一种高精度多粒度信誉计算模型，综合考虑相关推荐，排除恶意推荐和不可靠推荐的影响，提高信誉值计算精度，并根据应用需求进行多粒度的信誉评价。课题研究成果将为云计算的服务推荐等提供准确灵活的信誉评价机制。

云计算提供多种多样的服务类型和服务内容，使得用户在选择云服务时常常面临比以往更多的挑战，因此，建立一个客观可靠的信誉评价机制来选择可信服务对云计算推广非常重要。本课题针对云计算环境下服务的多样性和动态性、用户反馈的不完整性以及恶意评价的复杂性对信誉评价精度和粒度的挑战，研究面向云计算的高精度多粒度信誉评价机制，主要研究成果包括：1）对信誉评价机制进行理论分析，抽象出信誉评价机制的基本问题和形式化表示，分析影响信誉值计算精度的关键因素，作为后续研究的基础；2）提出一种基于用户特征的不完整评价填充方法，结合服务质量和用户评价特征预测用户缺失评价，和现有的不完整数据填充方法相比，我们的方法计算简单，适应于服务的动态特性，并且不依赖专家意见；3）设计一种高精度信誉计算模型，综合考虑用户对多个指标的评价关系来识别与过滤恶意评价和不可靠评价，引入多个调整因子（例如激励因子和自适应学习因子）来提高信誉计算的精度和适应能力；4）提出一种多粒度服务属性评价方法，挖掘用户对服务综合评价和指标评价的关系，自适应的反推出各个指标在综合信誉中的权重，组合相关指标信誉值从多种粒度对服务属性进行评价。5）在上述工作的基础上，研发面向云计算的信誉评价原型系统，分析和评估系统性能，对提出的方法和机制进行改进和优化。课题的关键技术成果发表在IEEE Transaction on Services Computing、IEEE Transaction on Vehicular Technology、Computer Communications、 AAAI、GlobeCom、TrustCom等期刊和会议上。