CAJ | 학술논문

针对现有推荐算法鲁棒性差的问题,提出一种融入个体特征差异的鲁棒协同过滤推荐算法.首先,根据用户评分信息的分布情况,给出用户评分个数偏离度和用户近邻平均相似度两个个体特征计算方法;然后基于真实用户和攻击用户个体特征的差异性,提出一种可疑用户标记算法;最后将可疑用户标记算法与矩阵分解技术相结合,对目标用户进行推荐.在MovieLens数据集上通过实验比较了提出的算法和其他相关算法的性能,实验结果表明算法不仅能够提高推荐精度,而且具有较强的鲁棒性.
침대현유추천산법로봉성차적문제,제출일충융입개체특정차이적로봉협동과려추천산법.수선,근거용호평분신식적분포정황,급출용호평분개수편리도화용호근린평균상사도량개개체특정계산방법;연후기우진실용호화공격용호개체특정적차이성,제출일충가의용호표기산법;최후장가의용호표기산법여구진분해기술상결합,대목표용호진행추천.재MovieLens수거집상통과실험비교료제출적산법화기타상관산법적성능,실험결과표명산법불부능구제고추천정도,이차구유교강적로봉성.
The existing recommendation algorithms have poor robustness against shilling attacks.In this con-sideration,in this paper we propose a robust recommendation algorithm incorporated with the difference of indi-vidual features.We first give two individual features which are user′s deviation degree of rating numbers and av-erage similarity of user′s neighbors.According to the distribution of users′ratings,we introduce the computation-al methods of individual features.Then we give the algorithm which can be used to label suspicious users based on the differences of computational results of individual features.Finally,we incorporate the matrix factorization technology with the identification results of suspicious users to make recommendations for users.Experimental re-sults show that the proposed algorithm not only improves the recommendation accuracy, but also has better ro-bustness.