Entropy-Guided Meta-Initialization Regularization for Few-Shot Text Classification (EGMI)

🎉 Our paper "Entropy-Guided Meta-Initialization Regularization for Few-Shot Text Classification (EGMI)" has been accepted for publication in Knowledge-Based Systems (Elsevier, SCI Q1, IF 8.8).

Requirements

Before running the code, ensure that you have all the necessary Python dependencies installed. The required libraries are listed in the requirements.txt file.

To install the dependencies, run:

pip install -r requirements.txt

This will install all the packages and libraries required to run the code.

Datasets

Provided Dataset:

• The HuffPost dataset is provided in the ./dataset folder by default.

Additional Datasets:

• You can also download other datasets for few-shot text classification via the following link.

Ensure you place the datasets in the appropriate folder as required by the scripts for proper loading during training.

Training & Testing

To train and test a BERT model using EGMI, follow these steps:

Training & Testing EGMI:

1. To train and test the model with EGMI, run the following command:

bash scripts/egmi.sh

This script will execute the training and evaluation for the EGMI method.

Training Standard MAML:

2. If you want to train the model using the MAML method (Meta-learning for few-shot classification), run:

bash scripts/maml.sh

This script will execute the training and evaluation using the standard MAML algorithm.

Summary:

• Install dependencies via pip install -r requirements.txt.
• Datasets are available in the ./dataset folder and additional ones can be downloaded from the provided link.
• For training with EGMI, use bash scripts/egmi.sh.
• For training with MAML, use bash scripts/maml.sh.