List your company
ResearchFormerly WizWand

Dissatisfied-turn Detection on human turn-level satisfaction annotations

36.55F1 ScoreUser-memory evaluator
0.763610.054319.34528.6357May 28, 2026
Updated 5mo ago

Evaluation results

MethodLinks
User-memory evaluatorType=User-aware LLM judge, Backbone=Qwen3-8B2026.0536.55
SPUR-style evaluatorType=Generic LLM judge, Backbone=Qwen3-8B2026.0529.55
Few-shot judgeType=Generic LLM judge, Backbone=Qwen3-8B2026.0526.8
Task-rubric judgeType=Generic LLM judge, Backbone=Qwen3-8B2026.0525.59
Nearest-history scoreType=User-history retrieval2026.0523.61
BERT scorerType=Supervised scorer, Backbone=BERT2026.0522.68
Prometheus-rubric judgeType=Generic LLM judge, Backbone=Qwen3-8B2026.0522.07
BERT ordinal scorerType=Supervised scorer, Backbone=BERT2026.0519.84
Zero-shot judgeType=Generic LLM judge, Backbone=Qwen3-8B2026.0517.32
RAG prompted scorerType=User-history retrieval2026.052.14