Adaptive Node Classification for Multiplex Graphs with Heterophily

A novel approach named \methodname has been introduced for classifying nodes in multiplex graphs, tackling the issue of heterophily, which occurs when connected nodes belong to different classes. Traditional models for multiplex graphs generally operate under the assumption of homophily, where connected nodes have similar characteristics, and they struggle with heterophilic relationships. Although recent techniques address heterophily in single-dimensional graphs, they do not capture the intricacies of multiplex graphs, where nodes are interconnected by various edge types that can display both homophilic and heterophilic interactions. \methodname employs dimension-specific compatibility matrices to represent different levels of homophily and heterophily across dimensions, enhancing the accuracy of node classification. This study is detailed in arXiv:2605.12699v1.