Utilizing Artificial Intelligence to Detect Fraudulent Manipulation in Recommender Systems

Abstract

The recommender system (RS) utilizes collective user opinions to forecast customer preferences. RS can be vulnerable to malicious information attacks. Introducing deceitful “shilling” profiles, such as push and nuke attacks, into the RS can promote inappropriate products. The introduction of these attacks results in inaccurate product recommendations. Because of their openness, Recommender systems are vulnerable to the injection of several fraudulent profiles, which might manipulate their predictions. Conventional detection methods rely on artificial characteristics derived from a single category of user-generated data. These methods need to comprehensively capture detailed user-item interactions, resulting in decreased accuracy when detecting diverse attacks. This study utilizes the modified density peak clustering algorithm to create a well-defined cluster using customer review information. The integration of collaborative filtering and content-based filtering methods has enabled a more advanced recurrent neural network algorithm to more precisely identify these hazards within hybrid recommendation systems. Characteristics obtained through varying degrees of corruption are ultimately incorporated into feeble classifiers capable of detecting attacks. Numerous assaults can be identified using the proposed method, as demonstrated by experimental testing on the Amazon dataset. In the realm of electronic commerce platforms, the recommended approach proves to be more effective for those experiencing rapid expansion in both product offerings and consumer base.