Snap KG’s success depends on graph link similarity
When Connected Does Not Mean Similar: Charting the Homophily Boundary of SNAP-KG for Streaming Entity Integration
Machine Learning
Summary
This paper looks at how a tool called SNAP-KG groups new pieces of information in growing knowledge maps based only on their own details, not their connections. The researchers found that SNAP-KG works well only when the linked parts of the map are similar, a quality called homophily. When the connected nodes are different (heterophily), SNAP-KG and similar tools struggle to group information correctly. The paper also suggests this problem is common for many similar methods and points to future directions for handling dissimilar connections.
What this means in practice
- •For knowledge graph engineers: Optimize entity grouping strategies by focusing on graph relations with similar connected nodes to improve clustering accuracy in dynamic knowledge graphs.
- •For data integration teams: Adjust entity integration workflows by identifying homophilous relationships to better assign new data entries without retraining models.
Authors
Jui-Chien Lin, Oshani Seneviratne
Abstract
SNAP-KG is a framework for assigning newly arriving entities to semantic communities in a growing knowledge graph (KG) using only their raw features, with no graph access and no retraining at inference time. It was evaluated on five multi-view benchmarks and a 2.4M-node OGB-WikiKG2 KG. In each of these datasets, at least one graph view is homophilous, meaning that connected nodes usually belong to the same class, and SNAP-KG performs well on all of them. This paper asks what happens outside that setting. We extend the evaluation to three heterophilous graphs (Texas, Wisconsin, Chameleon) and measure the edge homophily of every view. When no homophilous view is available, clustering quality drops sharply for both SNAP-KG and the transductive baselines used in its original evaluation. What decides this is the homophily of the relation, not the number of relations. Multi-view fusion still helps, but only when at least one homophilous relation provides a reliable foundation. The homophily assumption is therefore shared by the whole method family, not specific to SNAP-KG. We argue that heterophilous multi-view clustering is a separate research problem, outside the scope of this work. As future work, we outline how a heterophily-aware teacher could be distilled into SNAP-KG's projector to serve both homophilous and heterophilous KGs.