Factorized Hypothesis Search for Evidence-to-Taxonomy Retrieval
FHS is aimed at cases where the evidence does not directly name the taxonomy concept it should retrieve.
The paper defines that mismatch as a “retrieval readiness gap,” using examples like table cells whose meaning depends on surrounding row, column, datatype, and context. Its method keeps several partial interpretations across named semantic dimensions, then uses them for structured queries, multiple retrieval attempts, and candidate checks. On financial taxonomy tagging and CodiEsp clinical coding, it reports the best Recall@1, MRR, and final accuracy among non-oracle methods. HF Daily Papers' note
The paper defines that mismatch as a “retrieval readiness gap,” using examples like table cells whose meaning depends on surrounding row, column, datatype, and context. Its method keeps several partial interpretations across named semantic dimensions, then uses them for structured queries, multiple retrieval attempts, and candidate checks. On financial taxonomy tagging and CodiEsp clinical coding, it reports the best Recall@1, MRR, and final accuracy among non-oracle methods. HF Daily Papers' note
score 4