Uncertainty-aware graph neural network for semi-supervised diversified recommendation

Graphs are a powerful tool for representing structured and relational data in various domains, including social networks, knowledge graphs, and molecular structures. Semi-supervised learning on graphs has emerged as a promising approach to address real-world challenges and applications. In this paper, we propose an uncertainty-aware pseudo-label selection framework for promoting diversity learning in recommendation systems. Our approach harnesses the power of semi-supervised graph neural networks, utilizing both labeled and unlabeled data, to address data sparsity issues often encountered in real-world recommendation scenarios. Pseudo-labeling, a prevalent semi-supervised method, combats label scarcity by enhancing the training set with high-confidence pseudo-labels for unlabeled nodes, enabling self-training cycles for supervised models. By incorporating pseudo-labels selected based on the model’s uncertainty, our framework is designed to improve the model’s generalization and foster diverse recommendations. The main contributions of this paper include introducing the uncertainty-aware pseudo-label selection framework, providing a comprehensive description of the framework, and presenting an experimental evaluation comparing its performance against baseline methods in terms of recommendation quality and diversity. Our proposed method demonstrates the effectiveness of uncertainty-aware pseudo-label selection in enhancing the diversity of recommendation systems and delivering a more engaging, personalized, and diverse set of suggestions for users.