Overview

A content-based video game recommendation engine built as part of the INFO-H501 course. Designed end-to-end, the system ingests data from the IGDB (Internet Games Database) API, engineers multi-dimensional game features, and uses cosine similarity to surface relevant recommendations.

---

Methodology

1. Data Acquisition — Custom IGDB puller fetches game metadata, genres, platforms, ratings, community engagement metrics (follows, hypes), and time-to-beat data
2. Feature Engineering — Multi-label genre encoding via MultiLabelBinarizer, combined with standardized numerical features (rating, community engagement, playtime)
3. Similarity Computation — Cosine similarity matrix across the full feature space enables fast, precomputed recommendations
4. Interactive Interface — Streamlit application for exploring recommendations with fuzzy game name matching

---

How It Works

Given a game title, the system:

• Locates the game in the feature matrix (case-insensitive, with fuzzy matching fallback)
• Retrieves the top-N most similar games based on cosine similarity scores
• Returns recommendations with key metadata (rating, community follows)

The feature space blends categorical signals (genre combinations) with numerical signals (rating, engagement, completion time) to capture both thematic and quality-based similarity.

---

My Role

I designed the system end-to-end: the data ingestion pipeline from IGDB, the feature engineering approach, the similarity computation, and the Streamlit interface.

---

Tech Stack

• Python, Pandas, NumPy
• Scikit-learn (MultiLabelBinarizer, StandardScaler, cosine_similarity)
• Streamlit (interactive UI)
• IGDB API (game metadata)