HubScoreResult

HubScoreResult(
    scores,
    eigenvector,
    distance_to_center,
    eigenvector_z,
    distance_z,
)

Result of hub score computation.

Hub nodes are structural centers that hold the embedding space together. They are characterized by: - High eigenvector centrality (in tightly woven neighborhoods) - Low distance to global centroid (central position in space)

Unlike PageRank, which rewards being pointed to by many nodes (both categories AND popular instances qualify), eigenvector centrality rewards being in a tightly interconnected cluster. Abstract concepts cluster more tightly than popular instances.

Attributes

Name Type Description
scores np.ndarray Hub score for each point (higher = more structural/general)
eigenvector np.ndarray Eigenvector centrality for each point
distance_to_center np.ndarray Distance to global centroid for each point
eigenvector_z np.ndarray Z-scored eigenvector centrality
distance_z np.ndarray Z-scored distance to center

Methods

Name Description
get_hub_nodes Return indices of nodes with hub score above threshold.
get_top_hubs Return indices of top k hub nodes.

get_hub_nodes

HubScoreResult.get_hub_nodes(threshold=0.0)

Return indices of nodes with hub score above threshold.

get_top_hubs

HubScoreResult.get_top_hubs(k=100)

Return indices of top k hub nodes.