CAJ | 학술논문

协同过滤推荐系统中,推荐结果对用户偏好信息的敏感性使得推荐系统易受到人为攻击,即托攻击.恶意用户可以任意使用多重身份,或者是多个人来参与,都能注入恶意信息到推荐系统中.这类攻击严重影响了推荐系统的鲁棒性和准确性.这里深入分析了托攻击,结合主成分分析和变量选择方法,提出一个高精确度鲁棒的协同过滤系统架构,以保护推荐系统抵御用户概貌注入攻击.最后,通过实验验证表明该新型的高精确度的协同过滤系统可以取得更好的检测精度.
협동과려추천계통중,추천결과대용호편호신식적민감성사득추천계통역수도인위공격,즉탁공격.악의용호가이임의사용다중신빈,혹자시다개인래삼여,도능주입악의신식도추천계통중.저류공격엄중영향료추천계통적로봉성화준학성.저리심입분석료탁공격,결합주성분분석화변량선택방법,제출일개고정학도로봉적협동과려계통가구,이보호추천계통저어용호개모주입공격.최후,통과실험험증표명해신형적고정학도적협동과려계통가이취득경호적검측정도.
@@@@Collaborative filtering systems are vulnerable to manipulation by malicious social elements such as shilling attack. Malicious user votes and profiles could be injected into a collaborative recommender system, with either multiple identities or by involving more people, and this would significantly affect the robustness and accuracy of the system or algorithm, the shilling attack is analyzed in an in-depth way, and a novel robust collaborative filtering system, is proposed in combination with principal component analysis and variable selection, thus to protect recommender system from shilling attack. Finally the experiment indicates that this filtering system could achieve a fairly high detection accuracy.