rajdeep345 / ARRJuly2022

Submission for ARR July 2022

ARRJuly2022 Submission - CONTRASTE: A Multi-Task Approach Combined With Prompt-Based Contrastive Pre-Training For Aspect Sentiment Triplet Extraction

Dataset

• Preprocessed ASTE datasets can be found under 14res, 15res, 16res and lap14
• Preprocessed datasets for performing contrastive learning will be released upon acceptance of the paper.

Codes

• main.py contains the fine-tuning code for ASTE.
• utils.py contains helper codes for post-processing and evaluation.
• preprocessor.py contains codes to preprocess the ASTE datasets.
• Code for pre-training the model with supervised contrastive learning will be released upon acceptance of the paper.

Scripts

Contains the fine-tuning running scripts with different seed values and hyper-parameter configurations.