nesrinewagaa/Tunisian-Arabizi-Sentiment-Analysis

# Tunisian Arabizi Dialect Data - Sentiment Analysis ## Project Overview This project focuses on Sentiment Analysis for the Tunisian Arabizi dialect using multiple Deep Learning and Natural Language Processing (NLP) techniques. The objective is to classify Tunisian Arabizi text into sentiment categories such as: - Positive - Negative - Neutral The project explores several stages of the NLP pipeline including: - Data Exploration - Text Preprocessing - Data Augmentation - Tokenization & Sequence Preparation - Deep Learning Modeling - Hybrid Deep Learning Architectures - Transformer-Based NLP Models - Performance Evaluation & Visualization --- # Dataset The dataset used in this project is: **Tunisian Arabizi Dialect Data for Sentiment Analysis** The dataset contains: - Tunisian Arabizi textual data - Sentiment labels associated with each text sample Main columns: - `InputText` - `SentimentLabel` --- # Technologies & Libraries The project was implemented using Python and the following libraries: ## Data Processing - pandas - numpy ## Visualization - matplotlib - seaborn ## NLP - nltk - spaCy ## Machine Learning - scikit-learn - imbalanced-learn (SMOTE) ## Deep Learning - TensorFlow / Keras ## Transformer Models - Hugging Face Transformers --- # Project Pipeline ## 1. Data Exploration The dataset was explored using: - dataset shape inspection - missing value analysis - text length statistics - sentiment class distribution visualization ### Visualizations - Sentence length distribution - Sentiment label distribution --- # 2. Text Preprocessing Several preprocessing operations were applied: - Lowercasing text - Removing special characters - Cleaning punctuation - Sequence normalization --- # 3. Data Augmentation To improve model generalization, data augmentation techniques were applied. ## Synonym Replacement Words inside sentences were replaced using WordNet synonyms. ## Random Insertion Random synonyms were inserted into text samples to …