导　　师： 闵华清;毛祖铁

学科专业： 081202

授予学位： 硕士

作　　者： ;

机构地区： 华南理工大学

摘　　要： 随着系统应用范围的不断扩展,在商业利益的驱动下,为提升某些商品被推荐频率,商业竞争者开始频繁地在推荐系统中注入虚假评分记录,试图人为影响目标商品的推荐频率,扭曲推荐准确性与公正性,从而获取商业利润。而由于协同推荐系统具有天然的开放性,攻击者在掌握少量系统信息的前提下,即可成功对协同推荐系统推荐结果产生影响。这种行为影响了系统的准确性和公正性,因而引起了学界的注意。 本文首先阐述协同推荐基本原理和主流协同过滤推荐算法的算法流程,在此基础上,论文介绍现有协同过滤攻击的研究成果,并重点阐述攻击模型的形式化定义,提升性攻击模型的主要模式及组成结构,以及攻击模型对协同推荐算法所产生的影响。提出了一种基于用户信任的攻击防御模型,通过用户对系统推荐项目的反馈情况,与其邻居用户对该项目的推荐分进行分析评判,引入信任度机制,来降低攻击用户对推荐系统的影响。在实验中,我们使用了平均预测增量来评估基于信任的新模型对提升性的平均攻击和随机攻击的检测性能,大量的实验证明该模型能有效地降低协作过滤推荐中的随机攻击和平均攻击所产生的攻击效果,从而大大提高了推荐系统的可靠性。 With the expansion of the scope of application system, in the business, some goods are recommended for ascending frequency, commercial competitors began frequently in system with false record, trying to score goals for influence of commodities, distorted recommended recommend frequency accuracy and integrity, thus obtaining business profits. Due to the cooperative system has natural open, the attacker in mastering a system under the premise of information, can be successful in collaborative recommended recommend the system. This behavior affects the accuracy and integrity of the system, thus caused. This paper expounds the basic principles and main collaborative recommended cooperative filter algorithm recommendation algorithm, based on this, the paper introduces the current research collaborative filtering attacks against model, and focuses on the sexual assaults, ascension partition.formal definition model, and the model and structure of collaborative recommend against model algorithm. Is proposed based on users trust model, through the user attack on the system of feedback, recommended projects with the neighbors to users TuiJianFen analysis evaluation, introducing the trust mechanism, to reduce the system to recommend against users. In the experiment, we use the average increment to evaluate the new model based on trust to ascend the average attack and the detection performance of random attack, a large number of experiments shows that this model can effectively reduce the random collaborative-filtering recommend against and average attack produces effect, thus greatly improves the reliability of the system.

关 键 词： 协同过滤 推荐系统 攻击模型 信任度 电子商务

分 类 号： [TP301.6]

领　　域： [自动化与计算机技术] [自动化与计算机技术]