Table 3. Application of deep learning in ATE task.
| No | Study | DatasetDomain | Model | Performance | |||
|---|---|---|---|---|---|---|---|
| OE-acc | OE-F1 (%) | ACD-acc (%) | ACD-F1 (%) | ||||
| 1 | Poria, Cambria & Gelbukh (2016) | SemEval Res14 | CNN | – | 86.20 | – | – |
| CNN+LP1 | – | 87.17 | – | – | |||
| SemEval Lap14 | CNN | – | 81.06 | – | – | ||
| CNN+LP | – | 82.32 | – | – | |||
| 2 | Tran, Hoang & Huynh (2019) | SemEval Res14 | BiGRU-CRF | – | 85 | – | – |
| SemEval Lap14 | – | 78.5 | – | – | |||
| 3 | Xu et al. (2018) | SemEval Lap14 | DE-CNN | – | 81.59 | – | – |
| SemEval Res16 | – | 74.37 | – | – | |||
| 4 | Wu et al. (2018) | SemEval Res14 | unsupervised model | – | 76.15 | – | – |
| SemEval Lap14 | – | 60.75 | – | – | |||
| 5 | Ma et al. (2019) | SemEval Lap14 | Seq2Seq4ATE | – | 80.31 | – | – |
| SemEval Res16 | – | 75.14 | – | – | |||
| 6 | Liao et al. (2019b) | SemEval Res14-16 | LCC+GBC2 | 26.0% | 41.2 | – | – |
| SemEval Lap14, 16 | 33.7% | 36.1 | – | – | |||
| 7 | Xue et al. (2017) | SemEval Res14 | MTNA | – | 83.65 | – | 88.91 |
| SemEval Res15 | – | 67.73 | – | 65.97 | |||
| SemEval Res16 | – | 72.95 | – | 76.42 | |||
| 8 | Wu et al. (2019) | Digital QA reviews | MTA | – | 65.67 | 74.92 | 79.65 |
| Beauty QA reviews | – | 58.06 | 56.46 | 69.92 | |||
| Luggage QA reviews | – | 63.74 | 63.58 | 57.74 | |||
Notes:
1
LP: linguistic patterns.
2
LCC+GBC: model that coupling global and local context.