Steam Sentiment Analysis

Overview

This repository contains the notebooks, scripts, and research notes that power the Steam review sentiment experiments. It focuses on TF–IDF based classifiers, signal-rich exploratory visualizations, and documentation that explains how the NLP workflow evolved between 2019 and 2023.

Repository Layout

• game-review-sentiment-analysis.ipynb – primary notebook for feature engineering, model comparison, and chart generation.
• tfidf_model_tuner.py – reusable module for grid searching TF–IDF + logistic-regression pipelines and exporting coefficient charts.
• eda_analysis.py – utilities for dataset profiling and plot generation (review length histograms, sentiment vs. playtime, etc.).
• Sandbox/ – scratch notebooks that represent daily experiments referenced in the historical log.
• notebook_experiments.md / notebook_optimization.md – prose logs describing each exploratory run.

Environment Setup

1. Create and activate a virtual environment:

   python -m venv .venv
   .\.venv\Scripts\activate

2. Install the core dependencies:

   pip install jupyter pandas scikit-learn seaborn matplotlib nltk

3. (Optional) download the latest Steam review export into data/ if you want to reproduce the full training pipeline.

Running Workflows

• Launch notebooks:

  jupyter notebook game-review-sentiment-analysis.ipynb

• Run the TF–IDF tuner as a script to generate coefficient reports:

  python tfidf_model_tuner.py

• Emit updated exploratory charts:

  python eda_analysis.py

Quality & Automation

• Keep notebooks clean with jupyter nbconvert --ClearOutputPreprocessor.enabled=True before pushing changes.
• For code files, run python -m compileall . to catch syntax issues and format with ruff format or black (if available in your toolchain).

Exploratory Visualizations

• Added a stacked sentiment vs. review-volume chart to track weekend spikes.
• Logged genre-specific filters for the notebook demo to keep comparisons reproducible.
• Linked reports/visualizations/sentiment_volume.png for later sharing.

EDA Notebook Recipes

• Described how to call plot_sentiment_vs_hours() to generate stacked sentiment charts for stakeholder updates.
• Added reminder to store the resulting PNG in reports/visualizations/ for version control tracking.
• Highlighted the review segmentation filters we apply before exporting the chart.