Aspect-Based Sentiment Analysis (ABSA) 벤치마크
Aspect-Based Sentiment Analysis (ABSA) on ASQP
F1 (R15)
- 2021-08-01 — GAS: F1 (R15) 45.98
- 2021-10-02 — Paraphrase: F1 (R15) 46.93
- 2022-10-19 — DLO: F1 (R15) 48.18
- 2023-05-22 — MvP (multi-task): F1 (R15) 52.21
| Rank | Model | F1 (R15) | F1 (R16) | Paper | Code | Year |
|---|---|---|---|---|---|---|
| 21 | Gemma-3-27B (10-shot, self-consistency learning) | 39.95 | 46.23 | Do we still need Human Annotators? Prompting Large Language Models for Aspect Sentiment Quad Prediction | NilsHellwig/llm-prompting-asqp | 2025 |
| 22 | TAS-BRET | 34.78 | 43.71 | Aspect Sentiment Quad Prediction as Paraphrase Generation | isakzhang/absa-quad | 2021 |
| 23 | ChatGPT (gpt-3.5-turbo, few-shot) | 34.27 | – | MvP: Multi-view Prompting Improves Aspect Sentiment Tuple Prediction | ZubinGou/multi-view-prompting | 2023 |
| 24 | ChatGPT (gpt-3.5-turbo, zero-shot) | 22.87 | – | MvP: Multi-view Prompting Improves Aspect Sentiment Tuple Prediction | ZubinGou/multi-view-prompting | 2023 |