Adaptive pairwise judging improves language model evaluation accuracy
Backbone-Adaptive Evidence Routing for Robust Pairwise LLM Judging
Artificial Intelligence
Summary
Judging which of two language model answers is better can be tricky because different judging methods work better in different situations. The authors created a system called Backbone-Adaptive Evidence Routing (BAER) that changes how it compares answers based on the task, but makes sure its preferences stay fair if the two answers are swapped. They tested BAER on several benchmarks and found it was more accurate than other methods. This shows it helps to adapt judging methods rather than use the same one everywhere.
What this means in practice
- •For language model developers: Improve automated evaluation pipelines by selecting judging methods that adapt to different benchmarks for more accurate model comparisons.
- •For customer support bot teams: Enhance response quality evaluation systems to better select the best answer among multiple chatbot replies by using adaptive evidence gathering.
Authors
Zeyan Li, Jing Peng, Jianfeng Xu
Abstract
Pairwise language-model judges can gather evidence through direct comparison, reasoning, or reference-based verification, but no single protocol is best across benchmarks and judge backbones. We introduce Backbone-Adaptive Evidence Routing (BAER), which adapts the evidence mechanism while preserving candidate symmetry: swapping the two responses may reverse the preference but cannot change its strength. BAER separates each expert's signed preference from candidate-invariant reliability and builds three symmetric heads: evidence stacking, reliability-based expert routing, and candidate-blind reference verification. Development data select one head for each benchmark--backbone condition, and that choice is frozen before testing. Across four benchmarks and two 8B judge backbones, BAER achieves the highest test accuracy among the compared methods in all eight conditions, with full prediction coverage and gains of 0.87--7.32 points over the strongest external baseline. The results show that adapting how evidence is gathered is more reliable than fixing one judging protocol everywhere.