📱 Samsung Mobile Reviews – Sentiment Analysis Using Digikala API | Persian Text Processing | Machine Learning

⭐ About the Project

This repository contains a complete end-to-end sentiment analysis pipeline on Persian user reviews of Samsung mobile phones, collected directly from the Digikala API.

The project includes:

• Automated data collection from Digikala API (products + comments)
• Full data cleaning & preprocessing for Persian text
• TF-IDF vectorization optimized for short Persian reviews
• Handling imbalanced classes using targeted oversampling
• Training multiple ML models (Naive Bayes, Logistic Regression, Linear SVM)
• Evaluation using Accuracy, Precision, Recall, F1-score, and Confusion Matrix

📊 Dataset Overview

The dataset contains user reviews for Samsung mobile phones:

Column	Description
body	Original user comment
sentiment	Label assigned based on Digikala rating
comment	Preprocessed Persian text
label	Encoded label
(+ extracted product features from API in raw files)

🧠 Sentiment Classes

Reviews were categorized into:

• Positive
• Neutral
• Negative

Based on the numeric rating provided by users on Digikala.

🛠 Machine Learning Models Used

We trained and compared:

• Naive Bayes (Best Performance)
• Logistic Regression
• Linear SVM

Best model selected based on weighted F1-score and overall metrics.

📈 Evaluation Metrics

All models were evaluated using:

• Accuracy
• Precision
• Recall
• F1-score
• Confusion Matrix

Final results (example):

Model	Accuracy	F1-score
Naive Bayes	0.78	0.71
Logistic Regression	0.69	0.70
Linear SVM	0.73	0.71

🚀 How to Run the Project

1. Install dependencies: pip install -r requirements.txt
2. Run data collection (optional): python fetch_data.py
3. Open the notebook: jupyter notebook sentiment_analysis.ipynb

⚙ Preprocessing Techniques

• Persian text normalization
• Removing non-Persian characters
• Tokenization
• Stopword removal
• No stemming (better performance for short reviews)
• TF-IDF with n-grams (1,2)