基于多元隐式信任关系挖掘的抗攻击社会化推荐算法研究

在商业竞争环境下,推荐系统容易受到托攻击的危害.基于信任关系的社会化推荐算法被证明是解决托攻击问题的有效途径.然而,现有研究仅考虑显式信任关系,隐式信任关系没有被真正挖掘利用.为此,提出了一种基于多元隐式信任关系挖掘的抗攻击社会化推荐算法.首先,借鉴社会学和组织行为科学领域的信任前因框架模型,从全局信任和局部信任两个视角深入研究各信任要素的提取和量化方法.然后,通过信任调节因子集成局部信任度和全局信任度获得用户总体信任度.最后,以用户总体信任度为依据将攻击用户隔离在可信近邻之外,实现基于信任关系的个性化推荐.大量对比实验表明,本文算法在改善推荐准确率的同时有效抑制了托攻击对推荐算法的影响.

Anti-Attack Social Recommendation Algorithm Based on Multiple Implicit Trust Relationship

Under the commercial competition environment, recommendation system is vulnerable to shilling attack. The social recommendation algorithm based on trust relationship is proved to be an effective way to solve the shilling attack problem. However, most of the existing algorithms focus on explicit trust relationships, and implicit trust relationships are not really exploited. A social recommendation algorithm based on multiple implicit trust relation is proposed. Firstly, referring to the framework of trust antecedents in sociology and organizational behavior science, from the two perspectives of global trust and partial trust, we deeply study the extraction and quantification of trust elements. Then, the overall trust degree is obtained by integrating the local trust degree and the global trust degree by the trust adjustment factor. Finally, based on the overall trust of users, the attackers are isolated from trusted neighbors,realization of personalized recommendation. Extensive experiments are carried out on two datasets.The experimental results show that the algorithm improves the recommendation accuracy and effectively suppresses the impact of the shilling attacks.