Sentiment Analysis for Google Classroom Reviews

A lightweight, notebook-driven pipeline to scrape Google Play Store reviews for Google Classroom, preprocess English text, and run sentiment inference using two models: Logistic Regression (with TF‑IDF features) and a Dense Neural Network (DNN). The repo includes ready-to-run notebooks for scraping and inference, plus pretrained artifacts in the project root.

Features

• Scrape Google Play reviews and export to CSV (ulasan_classroom.csv).
• Robust text preprocessing: cleaning, lowercasing, slang normalization, tokenization, stopword removal, lemmatization.
• Dual-model predictions: Logistic Regression (TF‑IDF) and Dense Neural Network.
• Interactive, CLI-style input inside the inference notebook.
• Clear, reproducible environment via requirements.txt.

Tech Stack

• Python
• Data & NLP: NumPy, pandas, NLTK, scikit‑learn, joblib
• Deep Learning: TensorFlow/Keras
• Scraping: google‑play‑scraper
• Visualization (optional): matplotlib, seaborn, wordcloud
• Notebooks: Jupyter / VS Code Notebook

Key Files

• Notebooks:
  • Inference: Inference_Nelson_Ahli.ipynb
  • Scraping: Scraping_Nelson_Ahli.ipynb
• Models & vectorizer (pretrained):
  • TFIDF_model_sentiment.pkl
  • LR_model_sentiment.pkl
  • DNN_model_sentiment.h5
• Data sample:
  • ulasan_classroom.csv
• Project manifest:
  • requirements.txt

Installation

1. Create and activate a virtual environment (recommended)

macOS/Linux:

python3 -m venv .venv
source .venv/bin/activate

Windows (PowerShell):

python -m venv .venv
.\.venv\Scripts\Activate.ps1

2. Install dependencies

pip install -r requirements.txt

The pinned dependencies include (from requirements.txt):

• google_play_scraper, joblib, keras, matplotlib, nltk, numpy, pandas, protobuf, requests, scikit_learn, seaborn, tensorflow, wordcloud

3. Download NLTK resources (first run)

The inference notebook will attempt to download NLTK corpora at runtime. To pre-download:

python -c "import nltk; nltk.download('stopwords'); nltk.download('wordnet'); nltk.download('punkt')"

4. Ensure model artifacts are available

This repository already includes the pretrained artifacts in the project root:

• TFIDF_model_sentiment.pkl
• LR_model_sentiment.pkl
• DNN_model_sentiment.h5

If you move them, adjust paths in the inference notebook accordingly.

5. Adjust Colab-style paths (if needed)

Inside Inference_Nelson_Ahli.ipynb, the model/vectorizer are loaded from /content/... (a Google Colab path). For local runs, change these to relative paths, for example:

tfidf = joblib.load("TFIDF_model_sentiment.pkl")
model_lr = joblib.load("LR_model_sentiment.pkl")
model_dnn = tf.keras.models.load_model("DNN_model_sentiment.h5")

How to Run

Option A - VS Code Notebook

• Open Inference_Nelson_Ahli.ipynb in VS Code.
• Select the Python interpreter for your virtual environment.
• Run all cells (or step through them).
• At the input prompt, type a review (e.g., "this app is amazing") and view predictions from both models.

Option B - Jupyter (CLI)

python -m jupyter notebook
# or
python -m jupyter lab

• Open Inference_Nelson_Ahli.ipynb and run all cells.

Option C - Scrape dataset

• Open and run Scraping_Nelson_Ahli.ipynb.
• It uses google_play_scraper to fetch reviews for Google Classroom (com.google.android.apps.classroom) and saves to ulasan_classroom.csv.

Workflow Overview

1. Scraping

• Use Scraping_Nelson_Ahli.ipynb to fetch and save reviews into ulasan_classroom.csv.

2. Preprocessing & Inference

• The inference notebook:
  • Cleans and normalizes text
  • Replaces English slang using a CSV dictionary fetched from GitHub
  • Tokenizes, removes stopwords, and lemmatizes
  • Transforms with TF‑IDF
  • Predicts sentiment using Logistic Regression and DNN

3. Outputs

• Console output shows the predicted sentiment for each model.

Notes & Troubleshooting

• NLTK lookups: If you see lookup errors for stopwords, wordnet, or punkt, run the download commands in the Installation section.
• TensorFlow GPU: Not required; the DNN will run on CPU by default.
• Internet access: The slang dictionary is fetched over HTTP during preprocessing. If offline, cache it locally or replace with a local CSV.
• Path issues: If you keep artifacts in subfolders, ensure the notebook load paths point to the correct locations.

---

Maintained for the BFDL module - notebooks and models by Nelson Ahli.