对等计算环境中多维认知与动态推理的分布式信任机制研究

Peer-to-peer (P2P) is a distributed network architecture that provides resource sharing and collaboration between peers, and makes a significant impact on Internet Applications as one of the important components of future Internet. Open, anonymous, decentralized in nature and unrestricted environment of P2P architecture makes it an ideal environment for unauthorized access to resources and information, and hence it is difficult to implement security protections as compared to centrally administered systems. . Trust in P2P systems is a peer’s belief in another peer’s identity, reliability and capability based on its own experiences, can serve as the metric to decide the accessibility of peers to shared resources. That is meaningful to study distributed trust computing technologies for P2P networks, with reference value to other Internet-based self-organized networks, at the same time. . In this project proposal, we firstly present a P2P trust computing architecture, as a complementary component to a P2P system, to improve trust computational efficiency and flexibility with lower churning and topologies compatibility. . Based on this architecture, we adopt the historical behavior feedback to compute trust and propose multi-historical vector based evidence-modeling methods considering multiform of communication histories and diversities of users' trust requirements. Meanwhile, we design distributed evidence-secret sharing scheme to store evidence in P2P system. . Moreover, in order to recognize and resist various malicious behaviors with dynamic personality, we present multidimensional cognitive trust computing methods based on a trust/distrust dual model, including local cognition, global cognition, correlative cognition, personalized cognition and sparse cognition, with higher trust computing precision for dynamical malicious especially when sparse or no enough evidences exist. . Based on sociological balance theory and probability theory, we present trust inference algorithms for trust relations to predict malicious future behaviors and to improve computation capability for multi trust cognition models. Then, we propose algorithms to discover trusted evidence chains based on inference rules and present trust/distrust group discovery algorithms to predict spy malicious behaviors, and we also design inference based trust computing algorithms for multi-path complex trusted evidence chains. . At last, we propose a fuzzy trust decision model based on multi trust cognition and inference with Eigen factors of malicious behaviors. We also design a prototype system for simulations, verification and test during the course of this project.

对等计算网络对互联网起着积极的带动作用，是互联网应用的重要组成部分。但其对等、开放、匿名等特点带来了许多可信安全问题。研究信任计算技术对构建可信对等计算网络具有理论和实际意义，对其它基于互联网的自组织网络也具有参考价值。为提高信任计算效率及可扩展性，本项目首先设计具有松耦合接口的拓扑兼容的低抖动P2P信任计算架构。基于该架构，提出针对历史信息多样性、用户信任需求互异性的多维证据向量建模方法及分布式安全存储算法；基于多维证据、节点属性语义、网络关系提出多维认知的信任计算方法，以提高动态恶意环境下的信任计算精度与稀疏感知能力；利用社会学平衡理论与概率推理模型，提出复杂信任关系下的分布式信任推理算法，以进一步提高多维信任认知的不确定性计算能力及间谍恶意行为认知能力；最后设计区分多形态恶意行为的特征识别方法，结合多维认知与推理模型，提出综合信任认知的模糊决策模型及相关验证、推演原型系统。

对等计算网络对互联网起着积极的带动作用，是互联网应用的重要组成部分。但其对等、开放、匿名等特点带来了许多可信安全问题，如恶意篡改、恶意评价、共谋攻击等，影响了用户的信任决策能力。研究信任计算建模、预测、推理技术对构建可信对等计算网络具有理论和实际意义。本项目主要研究的内容包括拓扑兼容的低抖动P2P信任计算架构、多维信任证据向量的分布式建模与保密存储、基于证据向量多维认知的信任评估计算方法、基于节点网络关系与概率传播的动态信任推理方法等四个方面。项目组按计划完成了相关任务，搭建了原型系统实验平台，基于实际数据集进行了大量的科学研究。构建了低时间成本的迭代计算信任架构IDTRUST-ARCH，设计了分布式的三层通信架构及证据迭代计算框架，在时间成本上较传统计算模式较低了3个数量级；提出了面向信任证据分布式保护的多维球面分布式门限秘密算法，提高了信任证据分布式保护能力；在信任评估方面，基于节点个性化评价行为，提出了结合显性证据因子与隐性证据因子的多维信任建模方法，提出了多维局部信任度量模型、全局度量模型、信任决策模型、多云信任模型，可有效防御个体恶意攻击与团体共谋攻击；在信任预测方面，提出了解决冷启动、数据稀疏问题的基于比例模型的个性化评价预测算法、基于共振的个性化相似度计算模型，较大程度的提高了预测性能；在概率推理方面，基于物理学中的动力学模型提出了复杂网络关系下谣言行为传播概率模型，提高了行为预测的准确度。所做研究在IEEE ACCESS、IEEE Transactions on Knowledge and Data Engineering、软件学报、计算机研究与发展、CollaborateCOM、PST国内外期刊和会议上共发表标注论文21篇，其中SCI论文8篇（一区论文5篇），总影响因子达20.799；培养该领域内硕士毕业生10名。所做研究对提高对等计算行为分析、信任度量、信任决策能力具有参考意义，对云计算、社会计算中的节点信任度量、行为预测具有借鉴意义。