gpt4absa

Code for our paper: All in One: An Empirical Study of GPT for Few-Shot Aspect-Based Sentiment Anlaysis

Requirements

• Python 3.9
• PyTorch 1.12
• transformers
• openai
• tqdm

Stage 1 (Generate Candidates for ASTE and AOPE subtasks)

• Train a suitable model to generate candidates for the corresponding dataset as required by dual-encoder4aste;
• Please note that this step can be skipped, we have already generated suitable candidates in datasets.

Stage 2 (GPT inference)

• You can train the model using the corresponding .sh file ./run_with_ChatGPT.sh, ./run_with_GPTJ.sh or ./run_with_ERNIE.sh
• For example:

bash ./run_with_ChatGPT.sh

• Note: When using the GPT model, please add the correct openai.api_key and openai.api_base in ./model/model.py;
• Note：When using the ERNIE model, please add the correct API_KEY and SECRET_KEY in ./model/model.py.