Deep Dive into Compound E2E ABSA Tasks (top method)
As identified in issue 6: Identify ABSA single- or compound tasks to try to implement, this issue focusses on taking the next step: Digging deeper into the identified top methods.
The compound end 2 end tasks are especially relevant since they integrate both ATE (aspect term extraction) and ASC (aspect sentiment classification) into a single method, thereby capturing the dependencies between aspect terms and sentiments.
Here are (again) the results for top methods in E2E ABSA:
End 2 End Methods
- G. Yu, J. Li, L. Luo, Y. Meng, X. Ao, and Q. He, “Self question-answering: Aspect-based sentiment analysis by role flipped machine reading comprehension,” in Findings of EMNLP, 2021, pp. 1331–1342. [method 87]
- Y. Liang, F. Meng, J. Zhang, Y. Chen, J. Xu, and J. Zhou, “An iterative multi-knowledge transfer network for aspect-based sentiment analysis,” in Findings of EMNLP, 2021, pp. 1768–1780. [method 86]
- H. Luo, L. Ji, T. Li, D. Jiang, and N. Duan, “GRACE: Gradient harmonized and cascaded labeling for aspect-based sentiment analysis,” in Findings of EMNLP, 2020, pp. 54–64. [method 85]
Since Method 1. doesn't include any publication of code or their implementation, I will focus on methods 2. and 3.