Sentiment-Bert

Emotion Analysis Model (Arabic)

This repository contains a model for emotion analysis in Arabic text. The model predicts the emotion associated with a given text input. It was developed by Kirolos Amgad Khela and trained on the Emotone dataset and the SetFit/Emotion dataset after transcribing text to Arabic.

Model Details

• Model type: Text classification
• Language(s): Arabic
• Finetuned from model: MarBert

Usage

Direct Use

The model can be directly used for emotion analysis. It takes text inputs in Arabic and predicts the corresponding emotion associated with the text.

Training Details

Training Data

The model was trained using the Emotone dataset and the SetFit/Emotion dataset after translate text to Arabic.

Training Procedure

Preprocessing

The training data was preprocessed by cleaning the text from English and removing punctuations.

Training Hyperparameters

• Output Directory: "train"
• Logging Directory: "logs"
• Evaluation Strategy: "epoch"
• Per Device Train Batch Size: 32
• Per Device Eval Batch Size: 32
• Gradient Accumulation Steps: 1
• Number of Train Epochs: 3
• Learning Rate: 3e-5
• Warmup Ratio: 0.1
• Weight Decay: 0.01
• Adam Beta1: 0.9
• Adam Beta2: 0.999
• Adam Epsilon: 1e-8
• FP16: True
• Save Strategy: "epoch"
• Save Total Limit: 3
• Load Best Model at End: True
• Metric for Best Model: "macro_f1"
• Greater is Better: True

Dataset

The Emotone dataset and the SetFit/Emotion dataset are available for training. These datasets contain Arabic text annotated with emotion labels.

Fine-tuning in Colab

To fine-tune the model on a custom dataset, you can use the provided Colab notebook. Follow the steps outlined in the notebook to upload your dataset, configure the training parameters, and start the fine-tuning process.

Fine-tuned Model and Dataset

The fine-tuned model files and dataset are available in this Google Drive folder. You can download the necessary files from this folder.

Accuracy

The model achieved an accuracy of 82% on the evaluation dataset.

Authors

• Kirolos Amgad Khela

Contact

For any inquiries about the model, please contact Kirolos Amgad Khela at kirokhela5@gmail.com.