This repository contains the implementation of an emotion analysis model that predicts emotions from Arabic text. It leverages advanced machine learning techniques and NLP models like Arabert, Marabert, Gemini, and ChatGPT.
- Preprocessing of Arabic text data.
- Use of TF-IDF, BM-25, and Word2Vec for text vectorization.
- Integration of transfer learning models Arabert and Marabert.
- Emotion prediction using machine learning algorithms including KNN, SVM, and Random Forest.
- Utilization of generative AI models, such as Gemini and ChatGPT, to enhance emotion prediction capabilities.
- Python 3.x
- pandas
- numpy
- scikit-learn
- matplotlib
- seaborn
- nltk
- regex
- pyarabic
- emoji
- arabicstopwords
Clone the repository to your local machine:
git clone https://github.com/your-repo/emotion-prediction.git cd emotion-prediction
Install the required packages:
pip install -r requirements.txt
Load the data and run the main script:
import pandas as pd
df = pd.read_csv('your-data.csv')
from your_module import preprocess_data, predict_emotions
preprocessed_data = preprocess_data(df)
predicted_emotions = predict_emotions(preprocessed_data)
Word Representation | Algorithm | Accuracy | Precision (weighted avg) | Recall (weighted avg) | f1-score (weighted avg) |
---|---|---|---|---|---|
TF-IDF | KNN | 0.88 | 0.89 | 0.88 | 0.88 |
Random Forest | 0.88 | 0.89 | 0.88 | 0.88 | |
SVM | 0.89 | 0.90 | 0.89 | 0.89 | |
BM-25 | KNN | 0.81 | 0.84 | 0.81 | 0.82 |
Random Forest | 0.82 | 0.83 | 0.82 | 0.82 | |
SVM | 0.70 | 0.77 | 0.70 | 0.70 | |
Word2Vec | KNN | 0.69 | 0.73 | 0.69 | 0.71 |
Random Forest | 0.77 | 0.79 | 0.77 | 0.78 | |
SVM | 0.50 | 0.57 | 0.50 | 0.49 |
Model | Training Loss | Validation Loss | Accuracy | Roc Auc | f1-score (weighted avg) |
---|---|---|---|---|---|
UBC-NLP/MARBERTv2 | 0.08 | 0.25 | 0.66 | 0.90 | 0.83 |
CAMeL-Lab/bert-base-arabic-camelbert-mix | 0.09 | 0.24 | 0.64 | 0.89 | 0.82 |
aubmindlab/bert-base-arabertv2 | 0.14 | 0.25 | 0.61 | 0.87 | 0.80 |
Bert After Translation | 0.09 | 0.23 | 0.59 | 0.87 | 0.79 |
Model | Accuracy | ROC AUC | Precision | Recall | f1-score (weighted avg) |
---|---|---|---|---|---|
ChatGPT 3.5-turbo | 0.66 | 0.79 | 0.76 | 0.66 | 0.68 |
Gemini 1.5 Pro | 0.67 | 0.79 | 0.73 | 0.67 | 0.69 |
Contributions to this project are welcome! Please fork the repository and submit pull requests with your proposed changes.
Distributed under the MIT License. See LICENSE
for more information.
- Arabert and Marabert teams for providing pre-trained models.
- OpenAI for the ChatGPT model.
- Google for the Gemini model.