Most research on making graph neural networks robust has focused on undirected graphs, where edges are symmetric. A new arXiv paper argues that this focus misses a key signal: the trustworthiness encoded in the direction of links in directed graphs.
In a directed graph, an edge from one node to another is not interchangeable with an edge in the opposite direction. The authors contend that this directionality carries information about trust, and that existing robustness work has overlooked it.
The paper analyzes this gap and proposes redefining robustness for graph learning so that directed-graph trust is harnessed rather than ignored. The abstract is truncated, so the full method is not visible, but the central claim is clear: robustness research needs to take link direction seriously.