A Comparative Analysis of Bias Amplification in Graph Neural Network Approaches for Recommender Systems

• DOI: 10.3390/electronics11203301
• Publication Date: 2022-10-14T00:55:11Z
• Authors: Nikzad Chizari; Niloufar Shoeibi; María N. Moreno-G...
• Keywords: Information Overload; RSS

Recommender Systems (RSs) are used to provide users with personalized item recommendations and help them overcome the problem of information overload. Currently, recommendation methods based on deep learning gaining ground over traditional such as matrix factorization due their ability represent complex relationships between items incorporate additional information. The fact that these data have a graph structure greater capability Graph Neural Networks (GNNs) learn from structures has led successful incorporation into recommender systems. However, bias amplification issue needs be investigated while using algorithms. Bias results in unfair decisions, which can negatively affect company’s reputation financial status societal disappointment environmental harm. In this paper, we aim comprehensively study through literature review an analysis behavior against biases different GNN-based algorithms compared state-of-the-art methods. We also intend explore appropriate solutions tackle least possible impact model’s performance.