The Steam Dataset 2025: A Large-Scale, Multi-Modal Dataset of the Steam Gaming Platform

The Steam Dataset 2025: Complete Multi-Modal Gaming Platform Dataset

The Steam Dataset 2025 is a comprehensive, multi-modal dataset capturing the complete Steam gaming platform catalog as of August-September 2025. This dataset represents the largest and most methodologically rigorous Steam dataset available, containing 239,664 applications (games, DLC, software, and media) and 1,048,148 user reviews, with full preservation of nested data structures and pre-computed semantic embeddings.

The dataset is on Github and features full, detailed worklogs, all scripts and methodologies used in the production of the dataset

https://github.com/vintagedon/steam-dataset-2025

Key Features

Scale & Scope

239,664 total applications across all content types (games, DLC, software, tools, videos)

1,048,148 user reviews with full text and metadata

101,226 unique developers and 85,699 unique publishers

Coverage from 1997 to 2025 (28 years of platform history)

Data Purity

Exclusive use of official Steam Web API (no third-party data sources)

Complete preservation of nested JSONB structures (pricing, requirements, ratings)

Unmodified API responses maintaining data fidelity

Comprehensive metadata including release dates, pricing in 30+ currencies, and platform support

Multi-Modal Architecture

PostgreSQL 16.10 with full relational schema and JSONB support

Pre-computed 1024-dimensional BGE-M3 text embeddings for all applications and reviews

HNSW vector indexes enabling sub-second semantic search

Three data package formats (CSV, SQL dump, NumPy embeddings) for diverse use cases

Methodological Transparency

This dataset represents a novel approach to data publication by including complete project documentation:

Comprehensive Worklogs: Detailed session-by-session documentation of the entire data collection and processing journey, including challenges encountered, decisions made, and rationale behind technical choices

Jupyter Notebook Analyses: Pre-built analytical notebooks (exported as PDFs) demonstrating platform evolution analysis, semantic search implementation, genre clustering, and publisher network analysis

Complete Reproducibility: All data collection scripts, processing pipelines, validation procedures, and quality assurance steps documented and included

Educational Resource: First Steam dataset designed to serve dual purposes as both a research dataset and a learning resource for data engineering methodology

Schema Documentation: Comprehensive data dictionary, entity-relationship diagrams, and schema evolution documentation

This transparency enables researchers to understand not just the data, but the complete methodology used to create it, supporting reproducible science and serving as a reference implementation for similar large-scale data collection projects.

Materialized Features (Phase 2)

Platform support flags (Windows, Mac, Linux)

Pricing data (initial/final prices, discounts, currencies)

Achievement counts

Parsed PC system requirements (Phase 9)

Package Contents

This upload contains four distinct packages:

1. Accessibility Package (CSV) - 313 MB compressed

All tables exported as CSV files with UTF-8 encoding

Includes all materialized columns for immediate analysis

Ideal for: Students, data analysts, Excel/Pandas/R users

Excludes: JSONB columns, vector embeddings (available in other packages)

2. Power-User Package (SQL Dump) - 6.87 GB compressed

Complete PostgreSQL database dump with full schema

All JSONB nested structures preserved

All 1024-dimensional vector embeddings included

HNSW indexes for semantic search included

Ideal for: Researchers, data engineers, reproducible analysis

Requires: PostgreSQL 16+ with pgvector extension

3. AI Researcher Package (Embeddings) - 4.65 GB compressed

Pre-computed embeddings as memory-mappable .npy files

Mapping CSVs linking vectors to application/review IDs

Ideal for: ML engineers, AI researchers without GPU access

Enables: Direct loading into PyTorch/TensorFlow without recomputation

4. Repository Package (Complete Documentation) - 40.21 MB compressed

The complete GitHub repository export containing:

Comprehensive Worklogs: Session-by-session documentation of the entire project journey

Analytical Notebooks: Pre-built Jupyter notebooks (as PDFs) demonstrating:

Platform evolution and temporal growth analysis

Semantic game discovery using vector embeddings

Genre clustering and semantic fingerprinting

Additional analyses with sample data

Complete Documentation: README, data dictionary, dataset card, schema documentation

All Scripts: Data collection, processing, validation, and embedding generation code

Reproducibility Resources: Complete methodology documentation and quality assurance procedures

Ideal for: Researchers requiring full reproducibility, educators teaching data engineering, teams implementing similar data collection projects

Data Collection Methodology

Collection Period: August-September 2025

API Rate Limiting: Conservative 1.5s delays (17.3 requests/minute)

Success Rate: 56% (remaining failures due to delisted content, regional restrictions)

Embedding Model: BGE-M3 (1024 dimensions, multilingual support for 100+ languages)

Quality Assurance: Multi-phase validation with comprehensive documentation

Infrastructure: Proxmox Astronomy Lab enterprise platform with PostgreSQL 16.10 and NVIDIA A4000 GPU

Research Applications

This dataset enables research in:

Recommendation systems and collaborative filtering

Sentiment analysis and opinion mining (1M+ reviews)

Natural language processing on game descriptions

Semantic search and information retrieval

Market analysis and pricing strategies

Platform evolution and digital distribution studies

Graph analysis of publisher/developer networks

Time-series analysis of gaming trends (1997-2025)

Data engineering methodology and large-scale data collection workflows

Teaching reproducible science and transparent research practices

Schema Highlights

Core Tables:

applications (239,664 rows, 58 columns including vectors)

reviews (1,048,148 rows, 25 columns including vectors)

developers (101,226 unique entities)

publishers (85,699 unique entities)

genres (154 classifications)

categories (462 Steam feature tags)

Materialized Columns:

All Phase 2 pricing, platform, and achievement data extracted from JSONB

Vector Indexes: 9.8 GB of HNSW indexes (1.9 GB applications, 7.9 GB reviews)

Database Size: 21 GB total (data + indexes)

Technical Specifications

Database: PostgreSQL 16.10 (Ubuntu 16.10-1.pgdg24.04+1)

Extension: pgvector v0.8.0 for vector operations

Embedding Model: BAAI/bge-m3 (1024 dimensions)

Storage: Samsung PM983 1.92TB NVMe

Performance: ~205k TPS read-only, ~21k TPS durable writes

Citation

If you use this dataset in your research, please cite:

Donald Fountain. (2025). The Steam Dataset 2025: A Large-Scale, Multi-Modal Dataset of the Steam Gaming Platform. Zenodo. https://doi.org/10.5281/zenodo.17286923

Related Resources

GitHub Repository: https://github.com/vintagedon/steam-dataset-2025

Kaggle Dataset: [URL once published]

Documentation: Complete schema analysis, data dictionary, and methodology documentation included in repository

License

Creative Commons Attribution 4.0 International (CC BY 4.0)

You are free to:

Share: copy and redistribute the material

Adapt: remix, transform, and build upon the material

Under the following terms:

Attribution: You must give appropriate credit and indicate if changes were made

Version Information

Version: 1.0.0

Collection Date: August-September 2025

Last Updated: October 6, 2025

Database Snapshot Date: September 29, 2025