Advances in Graph Neural Networks and Recommendation Systems

The field of graph neural networks and recommendation systems is rapidly evolving, with a focus on improving robustness, generality, and efficiency. Recent developments have centered around addressing issues such as noise, data scarcity, and degree bias in graph-based models. Researchers are exploring novel approaches, including contrastive learning, optimal transport, and weak supervision, to enhance the performance and adaptability of these models. Notable papers in this area have demonstrated significant improvements in recommendation accuracy, node classification, and graph representation learning. Noteworthy papers include:

Heterogeneous Graph Masked Contrastive Learning for Robust Recommendation, which introduced a random masking strategy to enhance recommendation robustness.
HGOT: Self-supervised Heterogeneous Graph Neural Network with Optimal Transport, which proposed a novel self-supervised learning method without requiring graph augmentation strategies.

Sources

Heterogeneous Graph Masked Contrastive Learning for Robust Recommendation

Invariant Link Selector for Spatial-Temporal Out-of-Distribution Problem

Weak Supervision for Real World Graphs

Building a Recommendation System Using Amazon Product Co-Purchasing Network

HIEGNet: A Heterogenous Graph Neural Network Including the Immune Environment in Glomeruli Classification

HGOT: Self-supervised Heterogeneous Graph Neural Network with Optimal Transport

Learning Binarized Representations with Pseudo-positive Sample Enhancement for Efficient Graph Collaborative Filtering

Combining social relations and interaction data in Recommender System with Graph Convolution Collaborative Filtering

Multishot Capacity of Networks with Restricted Adversaries

Rethinking Contrastive Learning in Session-based Recommendation

Mitigating Degree Bias Adaptively with Hard-to-Learn Nodes in Graph Contrastive Learning