From Overall Sentiment to Aspect-Level Insights: A Pretraining Strategy for Unsupervised ABSA

Code and datasets of our paper "From Overall Sentiment to Aspect-Level Insights: A Pretraining Strategy for Unsupervised ABSA"

Requirements

• torch~=2.5.1
• transformers~=4.47.0
• pandas~=2.2.2
• tqdm~=4.66.5
• scikit-learn~=1.5.1
• numpy~=1.26.4
• evaluate~=0.4.3
• datasets~=2.19.1

Preparation

• This project uses the bert-base-uncased model from Hugging Face as the embedding layer for text.
• You can download train data from: https://www.dropbox.com/scl/fo/6faq3ormtc0p7qfi9yon1/AAbMHrP4VUcA6wF4og9HziQ?rlkey=c5fbsuksv3kah6ne2vxy316v3&st=9hojuloz&dl=0
• By default, the data is loaded from the folder: data/train/.
• If you'd like to use a different dataset, you can preprocess it using the script script/train_process.py before training.

ABSA trained model

To use the trained model for aspect-based sentiment analysis, you can download the model weights from the following link: https://www.dropbox.com/scl/fo/ekp4o1avws6tgycg8kpyc/AAFWG_ViaC-iNLbP9Y2JAWc?rlkey=ttsrb84rll23vat6txcwa4p5t&st=lkgtgfgh&dl=0

Once downloaded, save the weights to the specified folder. The default folder path is: model/trained_model Ensure that the model weights are placed correctly in the specified folder to allow the system to load them during analysis.

Run

To run the model, run:

sh run.sh