Med-aymen-alimi/Tunisian-Series-Sentiment-Analysis

Arabic sentiment analysis on YouTube comments of Tunisian TV series using machine learning and LSTM to understand audience opinions and feedback. # Tunisian Series Sentiment Analysis ## Project Overview Tunisian television series attract a large audience, particularly during the month of Ramadan. In recent years, many viewers have expressed dissatisfaction, claiming that producers focus more on controversy and media buzz than on meaningful and high-quality content. This project addresses this issue by applying sentiment analysis to YouTube comments related to Tunisian TV series. By analyzing public feedback, the project aims to extract valuable insights about audience opinions and help content creators better understand viewer expectations. The solution combines traditional machine learning algorithms with deep learning techniques to perform Arabic sentiment classification. --- ## Authors - Mohamed Aymen Alimi - Hedi Aloulou Supervised by: Mr. Wael Ouerda --- ## Project Structure ``` . ├── data.csv ├── sentiment_analysis.ipynb ├── Sentiment analysis.pdf │ ├── best_model.h5 ├── model_weights.h5 ├── model_architecture.pkl ├── model_config.pkl │ ├── best_model.pickle ├── best_model_sgd.pickle ├── vectorizer.pickle │ ├── deployment_lstm.py ├── deployment_sigmoid.py ``` --- ## Dataset Description - Source: YouTube comments related to Tunisian TV series - Total samples: 9,318 - Features: - Text: Arabic user comments - Sentiment: Target label ### Sentiment Labels - 1 : Positive - 0 : Neutral - -1 : Negative --- ## Methodology ### Data Extraction - Collected comments from selected YouTube videos related to Tunisian series - Retained only the comment text column - Removed emojis, numeric-only comments, repetitive words, and duplicates ### Data Annotation - Manual labeling of comments into: - Positive - Neutral - Negative ### Data Analysis - Exploratory analysis of sentiment distribution - Visualization using pie charts and bar charts ### Data Preprocessing - Removal of non-Arabic characters - Removal of HTML tags - Normalization of whitespace - Removal of single characters - Removal of null and dup …