融合随机森林和目标项目识别的鲁棒推荐算法

Robust recommendation algorithm combining random forest and target item identification

针对协同推荐算法面对托攻击时鲁棒性差的问题,提出一种融合随机森林和目标项目识别的鲁棒推荐算法。首先,基于卡方统计理论提取能够区分正常用户与攻击用户的有效特征。然后,训练随机森林分类器对攻击概貌进行第一阶段检测。接下来,通过识别目标项目对含有攻击概貌的类别做进一步检测,实现攻击概貌的第二阶段检测。最后,根据攻击概貌检测结果构建鲁棒推荐算法。实验结果表明,所提算法在保障推荐精度的前提下具有较强的鲁棒性。

The collaborative recommendation algorithms have lower robustness in the presence of shilling attacks. To address this problem, a robust recommendation algorithm combining random forest and target item identification is proposed. Firstly, chi-square statistics is used to extract the effective features which can distinguish normal users and attacking users. Then, the random forest classifier is used to classify user profiles, which is the first stage of the attack profile detection. Next, the further detection to the cluster including attack profiles is performed by identifying the target item, which is the second stage of the attack profile detection. Finally, based on the detection results of attack profiles, the corresponding robust recommendation algorithm is constructed. Experimental results show that the proposed algorithm outperforms the existing methods in terms of both recommendation accuracy and robustness.