Sentiment Analysis with Multi-Layer Perceptron (MLP)

This project demonstrates a complete machine learning pipeline for sentiment analysis using data scraped from the Google Play Store. The pipeline includes data scraping, exploratory data analysis, text preprocessing (TF-IDF), training various Multi-Layer Perceptron (MLP) classification models, optimizing hyperparameters, and saving the best model for inference.

Project Structure

• scrape_data.py - Script to scrape reviews from the Google Play Store and save them to a CSV file.
• scraped_reviews.csv - The dataset containing the scraped raw text reviews and their ratings.
• deep_learning_submission.ipynb - Jupyter Notebook encompassing the core workflow:
  • Data loading and cleaning
  • Exploratory Data Analysis (EDA)
  • Text Vectorization (TF-IDF)
  • Handling imbalanced data (SMOTE)
  • Training baseline, balanced, and optimized MLP models using scikit-learn
  • Evaluating models and plotting results
  • Saving the robust model
• inference.py - A command-line script to test the trained classification model on new text inputs.
• best_sentiment_model_final.pkl - The serialized form of the best performing NLP pipeline (TF-IDF + MLPClassifier).
• requirements.txt - Python dependencies needed to run the files.

Prerequisites

Make sure you have Python installed. You can install all necessary dependencies using:

pip install -r requirements.txt

Key Dependencies:

• google-play-scraper
• pandas, numpy
• scikit-learn, imbalanced-learn
• matplotlib, seaborn
• joblib

Usage Instructions

1. Data Scraping

If you want to gather new data from the Google Play Store, run the scraper script:

python scrape_data.py

This will produce or update the scraped_reviews.csv file.

2. Model Training

Open the Jupyter Notebook:

jupyter notebook deep_learning_submission.ipynb

Run all cells in the notebook. It will read the CSV dataset, train several permutations of the MLP model, evaluate their performance, and save the best model to best_sentiment_model_final.pkl.

3. Inference

You can test the trained model against various text inputs directly from your terminal using inference.py:

Testing predefined examples:

python inference.py

Testing a custom input:

python inference.py "Aplikasi ini sangat bagus dan bermanfaat"

The script will load the saved .pkl model and output the predicted sentiment class (Negative, Neutral, or Positive).