基于概率的信任传播模型  

Trust Propagation Based on Probability

作　　者：张绍武[1] 林鸿飞[1] 刘晓霞[1] 窦彦昭 

机构地区：[1]大连理工大学计算机科学与技术学院,大连116024

出　　处：《计算机科学》2014年第8期90-93,114,共5页Computer Science

基　　金：国家自然科学基金资助项目(60973068;61277370);辽宁省自然科学基金(201202031);教育部高等学校博士学科点专项科研基金资助课题(20090041110002)资助

摘　　要：社会网络中,用户之间的信任关系可以为用户判别信息是否可信提供依据。现有的信任计算方法一般是通过搜索节点之间的路径,再在其上添加各种其它限制,如路径长度、信任度下界等来计算信任度,而考虑节点之间的相似性的方法却很少。从节点之间的相似性出发,在信任传播模型的基础上,结合贝叶斯条件概率公式,提出了基于概率的信任传播模型。同时分析了信任传播模型中衰减系数对结果的影响;通过统计分析数据,得出具有信任关系的用户之间的相似度要比不具有信任关系的用户之间的相似度高得多,从而证明了贝叶斯理论可显著提高信任传播算法的有效性。在Epinion数据集上进行的实验证明了该方法的有效性。Trust relationship between users in a social network can provide ground to user to judge whether the infor- mation is trustworthy. Existing method of trust computes usually computes trust by searching the paths from source node to target node, and/or adding some external limitations, such as the length of path, the minimum trust values and so on. Few of them considers the similarity between nodes. This paper combined the trust propagation model with the similarity between nodes,and computed the distribution of the similarities between nodes by Bayesian probability for- mula. We also analyzed the influence of decay coefficient on the result and found the similarity between trust users is much higher than un-trust users in order to demonstrate the Bayesian probability formula can improve the method by data statistical analysis. At last, a trust propagation model based on probability was presented. After experiment on Epinion datasets,the validity of our method is proved by the result.

关 键 词：信任 社交网络 信任网络 信任预测 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]