🎮 Steam Review Sentiment Analysis with Metacritic Comparison | Machine Learning

This notebook performs sentiment analysis on Steam game reviews using three machine learning models:

• Multinomial Naive Bayes (Linear)
• Random Forest (Non-linear)
• SVM with RBF kernel (Non-linear)

It uses two datasets:

• Steam Reviews Dataset: A dataset containing raw reviews with language, game title, and a "recommended" label.
  • https://www.kaggle.com/datasets/najzeko/steam-reviews-2021/data
• Metacritic Dataset: A dataset with critic and user scores per game.
  • https://www.kaggle.com/datasets/thedevastator/video-game-ratings-and-reviews-dataset

🔄 Workflow Overview:

1. Load and input data from datasets using config.json
2. Clean and pre-process reviews (balance and filter reviews for model evaluation)
3. Vectorize text using TF-IDF
4. Train/test split + model training
5. Evaluate models (accuracy, precision, recall, F1, confusion matrix)
6. Compare model-predicted sentiment vs actual Metacritic scores (correlation + graphs)
7. Bonus insights:
   • Most common positive/negative words
   • Random review samples (with model correctness for error analysis)

⚙️ How to Use:

• No setup needed beyond having config.json and the datasets in your right file path
• Simply run each cell in order