一种结合用户可信度与相似度的鲁棒性推荐算法

Robust recommender algorithm based on user reliability and improved user similarity

协同过滤推荐系统面临着托攻击的安全威胁。研究抵御托攻击的鲁棒性推荐算法已成为一个迫切的课题。传统的鲁棒性推荐算法在算法稳定性与推荐准确度之间难以权衡。针对该问题,首先定义一种用户可信度指标;然后改进传统的相似度计算方法,通过结合用户可信度与改进的相似度,滤除攻击概貌,为目标用户作出推荐。实验表明,与传统算法相比,该算法具备更强的稳定性,同时保持了良好的推荐准确度。

Shillin gattacks pose a significant threat to the security of collabora tive filte rin grecom mender systems. It has cometo be an im po rtan t task to develop the attack-re sistant techniques fo r robust collab ora tive recom m endation. H o w e ve r, tra d itio nal collab ora tive filte rin g algorithm s have weakness in the balance between sta b ility and p re d ictive accuracy. To address thisp ro b le m , th is paper proposed user re lia b ility and im proved the ca lcu la tio n o f user s im ila rity , and incorporated both user re lia bility and s im ila rity in to standard collaborative filte rin g fram ew ork. E xperim ents show that the proposed a lgo rithm perform s betterthan state-of-the -art recom m ender algorithm s in s ta b ility and p re d ictive accuracy.