Topical Clusters in the AI Perception Graph
The AI Perception Graph organizes entities and topics into clusters. Cluster membership determines how AI systems assess topical authority — and whether a domain is treated as an expert source on a topic.
A topical cluster in the AI Perception Graph is a group of nodes (entities and topics) that are strongly interconnected through semantic relationships and appear across a coherent set of pages on the domain. Clusters represent the AI system's understanding of the domain's areas of expertise. A domain with strong, dense topical clusters is treated as an authority on those topics. A domain with weak, sparse, or disconnected clusters is treated as a generalist with no clear expertise.
How Clusters Are Formed
The Visualization Agent runs a graph clustering algorithm (specifically, a Louvain community detection variant) on the Perception Graph after construction. Nodes that are highly connected through perception edges form clusters. The cluster's primary entity is the node with the highest in-degree within the cluster — the entity that is most referenced by other entities in the cluster. The cluster is labeled with the primary entity's name.
Why Cluster Density Matters
Cluster density — the ratio of actual edges to possible edges within a cluster — is a topical authority signal. A high-density cluster means that every entity within the topic area is connected to multiple other entities within the same topic. This represents a comprehensive, interconnected treatment of the topic. A low-density cluster means the entities on a topic are isolated from each other — the domain mentions the topic but has not built an interconnected knowledge structure.
◆To build a high-density topical cluster: identify your primary entity for a topic, then ensure every supporting page explicitly names that primary entity, links to the hub page, and uses at least two other concepts from the same topic area. This creates the cross-references between supporting pages that make the cluster dense.
Gap Detection from Cluster Analysis
The Perception Graph analysis identifies cluster gaps: sub-topics or entities that would be expected in a comprehensive treatment of the domain's primary topic cluster but are absent. A machine trust intelligence platform would be expected to have perception nodes for trust scoring, entity analysis, citation analysis, retrieval simulation, and temporal authority. If any of these are absent from the perception graph, they represent coverage gaps that reduce the domain's perceived topical authority.
Perception vs Fact Layer — Part 9 of 10