WebKnoGraph: GNN framework optimizes internal links with 2x authority gain

WebKnoGraph is a new open-source framework from Emilija Gjorgjevska and colleagues that tackles a persistent SEO challenge: how to evaluate internal link changes before they go live. Traditional methods rely on manual judgment or fixed templates, making it hard to predict how a new link will redistribute authority across a site. The framework models a website as a directed graph, represents each page using embeddings, and uses the GraphSAGE algorithm to score candidate links. It then simulates interventions by embedding the site into a larger host graph (either a FineWeb-based real graph or a synthetic Barabási-Albert graph) and measures outcomes via PageRank-based authority metrics and semantic coherence scores.

The paper's key findings highlight a trade-off: automatic link selection consistently produces stronger authority redistribution (higher Authority Yield) but often degrades semantic coherence—meaning related pages become less connected. Expert-assisted selection, where human editors guide choices, better preserves coherence and, when targeting low-PageRank pages, actually achieves the highest Authority Yield. However, this comes with a less favorable loss-gain balance. The authors propose a practical workflow: generate candidate intervention sets at scale, evaluate them jointly across authority gain, volatility, loss-gain balance, and semantic coherence, then review for editorial deployability. This hybrid approach offers SEO professionals a data-driven way to optimize internal linking without blindly sacrificing site structure.