PKDDialogue.TurnSemanticAnalyzer.GetTopNMatchingTurnsByDecompositionRelevanceToDocument Function
Instead of creating a basis by projecting words in the query, we create a basis by projecting words
in each document. Then we project the query into that basis. So this is more of a backwards looking
rhetorical relationship. We expect the question to be highly relevant to the answer, and not add any
information in particular to the answer.