Association probes query models unprompted across three distinct directional prompts to map brand recall, normalising the extracted names through a prefix-based Matryoshka attribution hierarchy.
flowchart LR
subgraph Modes[Probing Modes]
Q[Query Text] -->|Q2B| B1[Brand Names]
E[Entity Topic] -->|E2B| B2[Brand Names]
BR[Brand Name] -->|B2E| C[Associated Concepts]
end
subgraph Norm[Normalisation]
B1 --> N[Strip Noise]
B2 --> N
N --> M[Matryoshka Prefix Tree]
M --> R[Root Brand Aggregate]
endReferenced by