Sentiment Sextuple Extraction on PanoSent 1.0 (val)
44.37Avg F1 (Sextuple Micro/ID)MSGR+SFT-Qwen3-8B+HLOS
Evaluation results
| Method | Links | |
|---|---|---|
| MSGR+SFT-Qwen3-8B+HLOSModules=MSGR + HLOS, Backbone=SFT-Qwen3-8B2025.12 | 44.37 | |
| SFT-Qwen3-8B+HLOSModule=Hybrid LLM Optimization Strategy (HLOS), Backbone=SFT-Qwen3-8B2025.12 | 43.59 | |
| MSGR+SFT-Qwen3-8BModule=Multi-Sampling Generation and Refinement (MSGR), Backbone=SFT-Qwen3-8B2025.12 | 37.24 | |
| SFT-Qwen3-8BTraining=Supervised Fine-Tuning, Backbone=Qwen3-8B2025.12 | 35.61 | |
| Sentica (+CoS+PpV)Prompting Strategy=Chain-of-Symbol + PpV2025.12 | 33.95 | |
| Sentica (+CoS)Prompting Strategy=Chain-of-Symbol2025.12 | 31.7 | |
| Sentica (+CoT)Prompting Strategy=Chain-of-Thought2025.12 | 28.82 |