Abstract
dc:description.abstract<p>We propose a simple yet effective regularization technique for node classification on graphs that leverages edge-based label co-occurrence patterns. We first train an MLP on node features to produce class probability distributions, then compute a fixed penalty matrix from edge-based co-occurrence statistics of these predictions. This penalty matrix, which captures unlikely class combinations on connected nodes, is then used to regularize GNN training without further updates. We evaluate this approach across multiple homophilic datasets (Cora, CiteSeer, PubMed, ogbn-arxiv) and heterophilic benchmarks (Chameleon, Squirrel, Actor, Roman-Empire) using three GNN architectures: GCN, GraphSAGE, and H2GCN. Results show consistent improvements on homophilic graphs, with accuracy gains of 3-6% on Cora and CiteSeer. The method shows more modest gains on heterophilic datasets, suggesting its effectiveness is tied to graph structure. Our experiments demonstrate that simple edge co-occurrence statistics from feature-based predictions can enhance GNN node classification without architectural modifications.</p>
Degree
thesis:*- Name thesis:degree_name
- Master of Science in Computer Science - (M.S.)
- Discipline thesis:degree_discipline
- Computer Science
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Altunel, Kadir
- Contributors dc:contributor
-
- Ioannis Koutis
- Shivvrat Arya
- Mengjia Xu
Subjects
dc:subject × 11Identifiers
dc:identifier.*- Repository record dc:identifier
- https://digitalcommons.njit.edu/theses/3545