Nodes & edges

Nodes are concepts

Each node is an entity GraphRanks extracted from the content — a meaningful concept such as a product, a technique, or a topic. Surface forms are normalised to a single canonical concept, so “knowledge graphs” and “knowledge graph” become one node rather than two. Every node carries a type (Product, Concept, Organization, and so on) shown in its label and colour.

Edges are co-occurrence

An edge connects two concepts that appear close together in the source text. GraphRanks slides a small window across the content and links concepts that fall inside it, so an edge means “these ideas are discussed together,” not a hand-authored fact. The relationship is labelled co_occurs_with.

Each edge has a weight — how often the two concepts co-occurred, summed across the content. Heavier edges pull their concepts together in the layout, so tightly related ideas form visible clusters.

Reading the graph

• Clusters are groups of densely connected concepts — the sub-topics that make up the subject.
• Degree is how many neighbours a node has. High-degree nodes are the hubs the topic revolves around.
• Disconnected nodes appear in the content but link to nothing else — often a mention in passing rather than real coverage.
• Weakly connected nodes hang off the graph by a single edge — concepts touched on but not developed.

The overview strip at the top of a run summarises these counts (entities, relations, disconnected, weakly connected, clusters) so you can gauge a graph's shape at a glance.