基于改进支持向量机的推荐系统托攻击检测方法

摘要

A support vector machine algorithm with good generalization ability is used to establish the shilling attack detec⁃tion model of recommendation system. In traditional support vector machine algorithm,the specific parameters of the penalty fac⁃tor used to control the misrecognition sample penalty and insensitive loss parameters are determined by users,which can decide the performance of support vector machine to a great extent. The convergence performance of the standard PSO algorithm de⁃pends on the selection of the important parameters,such as learning operator and inertia coefficient. The convergence speed of the standard PSO algorithm is fast at early stage and relatively slow at latter stage,and the algorithm is easy to fall into local minimum (early maturity) due to PSO convergence. Therefore,both the chaos optimization algorithm and PSO algorithm are used to optimize the traditional support vector machine algorithm. The MovieLense100K dataset is used for instance analysis. The detection results show that the higher the filling rate is,the higher detection accuracy becomes. The improved support vec⁃tor machine algorithm has the best detection performance,and can help the recommendation system to prevent the shilling at⁃tack,so as to obtain more accurate user rating data.%使用具有较好泛化能力的支持向量机算法建立推荐系统托攻击检测模型，由于在传统支持向量机算法中，用来控制错误识别样本惩罚度的惩罚因子的具体参数以及不敏感损失参数的具体参数由使用者决策，并在较大程度上决定支持向量机的性能。标准PSO算法的收敛性能基本取决于学习算子和惯性系数等重要参数的选取。标准PSO算法前期收敛速度很快，后期则比较缓慢，粒子群趋同性造成算法后期容易陷入局部最小值，即进入早熟。因此，使用混沌优化算法与PSO算法共同完成对传统支持向量机算法的优化。最后使用MovieLense100K数据集进行实例分析，从检测结果对比可以看出，填充率越高，检测准确率越高，研究的改进支持向量机具有最优的检测性能，能够帮助推荐系统防范托攻击，以得到较精准的用户评分数据。