講演情報
[6F-04]Collaborative Filtering Recommender System against Attacks based on Dual Clustering
*Fu Lei1、張 建偉1 (1. 岩手大学 理工学研究科)
発表者区分：学生
論文種別：ロングペーパー
インタラクティブ発表：なし
論文種別：ロングペーパー
インタラクティブ発表：なし
キーワード：
Recommender system、Collaborative filtering、Clustering、Shilling attack
Collaborative filtering has emerged as a widely adopted solution to address the challenge of information overload in recent years. However, it remains vulnerable to shilling attacks, wherein attackers introduce fake ratings to manipulate recommender systems. This paper proposes a collaborative filtering method designed to mitigate the impact of such attacks while maintaining prediction accuracy by employing a dual clustering method on the dataset. The dual clustering process involves two stages: the first clustering partitions the dataset into multiple clusters, while the second clustering integrates these clusters to form less clusters. This method is intended to separate actual users from attack users into different clusters by the first clustering and to increase the data volume of actual user data within one cluster by the second clustering. The effectiveness of the proposed dual clustering method is validated through experiments conducted on a large movie ratings dataset. The prediction accuracy is evaluated by measuring the error between actual and predicted ratings, while robustness against attacks is assessed by examining the shift of TopN value before and after the attacks. Experimental results indicate that the dual clustering method effectively reduces the impact of the drawbacks in collaborative filtering, demonstrating its potential to enhance both prediction accuracy and robustness against attacks.