Addressing Bias in Recommendation Systems: A Debiased Deep Cross Network Approach
Keywords: personalized recommendation systems, bias, detection, mitigation, eigenvector centrality, influential nodes, regularization technique
TL;DR: Detecting and mitigating bias in recommendation systems
Abstract: With the increasing prevalence of personalization of recommendation systems in various domains, this research paper examines the presence of bias associated with such systems. The study focuses on the detection and mitigation of bias through the development of a baseline hybrid model that predicts ratings and generates user recommendations. Bias detection is achieved by analyzing disparities in the model's performance across different population subgroups, while the calculation of the eigenvector centrality of nodes aids in identifying influential nodes within the recommendation system's network. To mitigate bias, a regularization technique is employed,adjusting the impact of user ratings based on movie popularity. The effectiveness of the regularization technique is demonstrated by the low root mean squared error (RMSE) scores, highlighting its success in addressing bias within personalized recommendation systems.
Submission Category: Machine learning algorithms
Submission Number: 38
Loading