A convex optimization method that projects instructor or LLM-judge scores onto the space of proper textual scoring rules, producing scores that fit the reference better than prior proper baselines.
Split and Rewrite each evaluative statement into several abstract and concise statements, abandoning the specific expression
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Aligned Textual Scoring Rules
A convex optimization method that projects instructor or LLM-judge scores onto the space of proper textual scoring rules, producing scores that fit the reference better than prior proper baselines.