Bias in Recommender Systems: Item Price Perspective

Recommender systems are a widely studied application area of machine learning for businesses, particularly in the e-commerce domain. These systems play a critical role in identifying relevant products for customers based on their interests, but they are not without their challenges. One such challenge is the presence of bias in recommender systems, which can significantly impact the quality of the recommendations received by users. Algorithmic bias and popularity-based bias are two types of bias that have been extensively studied in the literature, and various debiasing methods have been proposed to mitigate their effects. However, there is still a need to investigate the mitigation of item popularity bias using product-related attributes. Specifically, this research aims to explore whether the utilization of price popularity can help reduce the popularity bias in recommender systems. To accomplish this goal, we propose mitigation approaches that adjust the implicit feedback rating in the dataset. We then conduct an extensive analysis on the modified implicit ratings using a real-world e-commerce dataset to evaluate the effectiveness of our debiasing approaches. Our experiments show that our methods are able to reduce the average popularity and average price popularity of recommended items while only slightly affecting the performance of the recommender model.