Keymaker Evaluates Questions By:
-
Relevance: Measures how directly a question supports the explicit goal of the current step, ensuring tight alignment with the process trajectory.
-
Predicted Information Gain: Quantifies the potential increase in meaningful knowledge if the question were answered, estimated by comparing semantic novelty against Keymaker’s current memory graph density and existing concept coverage.
-
Uncertainty Reduction: Evaluates how effectively the question targets low-confidence nodes or contested regions within the Map of Meaning (MoM), with emphasis on resolving ambiguity.
-
Domain-Specific Priority Boost: CF (Cystic Fibrosis) and 2184insA mutation-related topics are prioritized with a 20% score amplification to reflect strategic research emphasis.
-
Confidence Anchor: Gives preference to questions referencing entities or concepts with internal confidence scores below 0.85, ensuring attention is given to areas needing higher validation or clarification.
Scoring Function (Initial):
Score = (0.4 * Relevance) + (0.3 * Predicted Info Gain) + (0.2 * Uncertainty Reduction) + (0.1 * Domain Boost)
The question with the highest composite score is selected for answer generation.