利用离群点算法预处理协同过滤推荐系统数据

摘要

由于电子商务系统的开放性和推荐系统用户的广泛参与性，推荐系统很容易受到攻击。出于某种目的的用户向系统中注入恶意信息，导致推荐质量受到威胁，因此过滤掉恶意信息成为迫切需要。离群点检测用于从数据集中找到明显偏离其他数据对象或不满足一般对象行为特征的对象。为了提高推荐系统的鲁棒性，保证推荐系统的高质量，文中利用局部离群点检测算法计算出每个用户的局部离群因子( LOF)，过滤掉离群因子较高的用户，然后运用协同过滤算法为系统中剩下的用户做推荐。实验结果表明，与传统的协同过滤推荐算法相比，此方法在提高推荐质量上取得了一些好的效果。%Due to the openness of the e-commerce system and extensive participation of recommended system users,recommendation sys-tem is vulnerable to attack. Some users who want to reach a particular purpose inject malicious information into the system,leading to un-der threat of the recommendation quality,and thus it is necessary to filter out malicious information. Outlier detection is to find the excep-tional objects which do not satisfy the common patterns or deviate much from the rest objects of the dataset by some measure. In order to improve the robustness and guarantee the high quality of the system,compute user’ s Local Outlier Factor ( LOF) and remove users who has a higher local outlier factor based on local outliers algorithm,and then use the collaborative filtering algorithm to recommend for the users. Compared with the traditional collaborative filtering algorithm,the experimental result shows some good results have been achieved on improving the quality of recommendation.