Measure the graph.
Understand the questions.
Analyze knowledge-graph structure, QA collections, and joint KGQA navigation difficulty for Kinship, MQuAKE-ST, MetaQA, or your own files. Everything runs locally in the browser.
KG, QA, and joint statistics on demand.
Select a graph, a QA collection, or both. The report reveals graph topology, QA composition, and joint navigation difficulty measures.
Statistics summary
Consistent definitions across KG, QA, and joint views
Duplicate triples are counted in input rows, then removed for navigation-based degree and random-walk calculations. Weak components and density use a simple undirected projection. QA summaries report the selected rows after any split filter.
Joint analysis uses the selected graph with directed or undirected traversal actions, released source and answer entities, and reference-path annotations when available. Fixed-horizon random-walk probability evaluation is optional. Graph structural metrics always use the original triples.
Your files stay in your tab
Custom files are processed locally in a Web Worker and are not uploaded. Built-in resources are downloaded from Hugging Face or the Google Cloud Storage JSON API and cached in this tab until cleared or the page closes or reloads. Large graphs may take time and memory. Optional expensive calculations have size caps and are labeled as skipped when necessary.
When only QA is selected, KG-only tabs explain that no graph was loaded. When both are selected, the joint report combines them in a single client-side run.