Literature Review on Aspect and Sentiment Extraction

[fani-lab]

Literature Review on Aspect and Sentiment Extraction

[farinamhz]

As proposed by the authors in CONTRASTE: Supervised Contrastive Pre-training With Aspect-based Prompts For Aspect Sentiment Triplet Extraction, current leading methods can be categorized mainly into five distinct groups in which I think generative ones are great for the case of latent aspect detection with no surface form:

1. Pipeline
2. Tagging-based
3. Span-based
4. Generative
5. MRC-based

Among generative ones, one group of the methods is based on BERT LM, and the other group is based on T5 LM.

*This paper has been published in EMNLP'23 but has not become available yet, so I provided the link from Arxiv.