Emotion-Classification-using-LLMs

Contributors

Created by Marc-Antoine Nadeau, Jessie Kurtz, and Baicheng Peng

Overview

We investigate and compare the performance of the following models in emotion classification for the GoEmotion dataset:

Large Language Models (LLMs):

• BERT
• GPT2
• Distilled-GPT2

Traditional Models:

• Naive Bayes
• RF
• SR
• XGBoost

GoEmotions

The GoEmotions dataset, developed by Google Research, contains over 58,000 English Reddit comments, each labeled with one or more of 27 emotion categories or marked as neutral. This project focuses on classifying these emotions and evaluating performance using several algorithms and configurations.

File Structure

How to Run

BERT

1. Navigate to the src/BERT folder.
2. Run the PrepareDataset.py file.

Distilled-GPT2

1. Navigate to the src/Distilled-GPT2 folder.
2. Run the PrepareDataset.py file.

Naive Bayes

1. Navigate to the src/Naive-Bayes folder.
2. Run the PrepareDatasetNB.py file.

Baseline Model

1. Navigate to the src/Baseline_model folder.
2. Run the RandomF.py file.

GPT2

1. Navigate to the src/GPT2 folder.
2. Run the PrepareDataset.py file.

Word2Vec

1. Navigate to the src/Word2Vec folder.
2. Run the word2vec.py file.

Results

Any resulting figures, graphs, and CSV will be saved in their respective folders:

• Results-BERT: Contains the results for the BERT LLM.
• Results-Distilled-GPT2: Contains the results for the Distilled-GPT2 LLM.
• Results-GPT2: Contains the results for the Distilled-GPT2 LLM.
• Results-Naive-Bayes: Contains the results for the Naive-Bayes.
• Results-Word2Vec: Contains the results for the Word2Vec.
• Results-RandomF: Contains the results for Baseline Model (RF)

Prerequisites

Make sure you have the following Python libraries installed:

• numpy
• matplotlib
• torch
• seaborn
• pandas
• transformers
• scikit-learn

You can install them using pip:

pip install numpy matplotlib torch seaborn pandas transformers scikit-learn