CritiqueMeta

CritiqueMeta collects video-game rating data and explores the biases between professional critics and players: are users getting tougher? are ratings constantly increasing? what drives user dissatisfaction?

v2 (2026) — the project was rebuilt around Metacritic's JSON API (the HTML scraper died with the 2023 site redesign) and the analysis moved from a static notebook export to an interactive Streamlit app. Coverage grew from 7 consoles (2012–2023) to the entire Metacritic catalog — every platform, PC included.

Architecture

critiquemeta/          collection & preparation CLIs
  mc_client.py         Metacritic API client (rate-limited, retries)
  scrape.py            data collection, incremental & resumable
  clean.py             raw JSONL -> analysis-ready parquet/CSV
  opencritic.py        optional OpenCritic comparison sample (RapidAPI key)
app/                   Streamlit app (one page per theme)
raw_data/              scraping checkpoints (JSONL, git-ignored) + legacy 2023 CSVs
data/                  cleaned datasets (parquet)
legacy/                v1 notebooks and the old HTML report

Setup

python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

1 — Collect

All commands are incremental: interrupt them freely, re-run to resume.

python -m critiquemeta.scrape catalog        # full per-platform catalog (~500k rows, ~1h)
python -m critiquemeta.scrape details        # developer/publisher for every rated title
python -m critiquemeta.scrape user-stats     # user rating counts per (title, platform)
python -m critiquemeta.scrape reviews-critic # individual critic reviews (long!)
python -m critiquemeta.scrape reviews-user   # individual user reviews, capped per game
python -m critiquemeta.scrape status         # progress overview

Useful flags: --platform playstation-5 (restrict), --limit N (smoke test), --rate / --workers (throughput, keep it polite), --max-per-game / --full (user review depth).

The collection uses backend.metacritic.com, the JSON API behind metacritic.com, with the site's own public API key. Intended for personal / research use — keep the default rate limits reasonable.

2 — Clean

python -m critiquemeta.clean

Produces data/games.parquet (one row per title-platform, with derived columns: offset = Metascore − user score, ratio = user ratings / critic reviews, …), data/meta_reviews.parquet and data/user_reviews.parquet.

3 — Explore

streamlit run app/Accueil.py

Pages: score distributions, platform view, game-by-game scatter & explorer, critic-publication biases, developers, time trends, and an optional Metacritic-vs-OpenCritic comparison.

Optional — OpenCritic

Create a free key on RapidAPI (OpenCritic API), then:

export RAPIDAPI_KEY=...
python -m critiquemeta.opencritic --top 300

The free quota is small; the script checkpoints and resumes across days. Once data/opencritic.csv exists, the OpenCritic page of the app lights up.

Deploying the app

The repo is directly deployable on Streamlit Community Cloud: point it at app/Accueil.py. The parquet datasets in data/ are versioned so the cloud app works without re-scraping.