FAQ
SteamTLDR is a personal project and isn't affiliated with Valve or Steam. Game names, images and reviews belong to their respective owners.
Why SteamTLDR?
Steam gives you one score, so finding out why players love or hate a game means reading dozens of reviews or watching a video, and a single number hides what they praise, what they complain about, and whether opinion is changing. SteamTLDR reads thousands of reviews per game and sums up the why: top themes, recent trends, and how new players compare with veterans.
Data sources
- Reviews come from Steam's public review API, English only.
- Scores, monthly trends and the playtime split are Steam's exact totals, not estimates.
- The text sample per game: the 1,000 most helpful reviews, 1,000 from the last 30 days, and up to 4,000 spread across earlier months.
- Each game is a snapshot, and its page shows when it was last updated.
Finding themes
- Reviews are split into clauses at punctuation and turning-point words like but and however, so "great story but constant crashes" is praise for the story and a complaint about crashes.
- Each clause is matched against 15 themes, each a hand-picked keyword list.
- A clause takes its review's thumbs up or down, unless the sentiment model is confident it's the opposite (below 40% positive in a thumbs-up review, above 80% in a thumbs-down one).
- Bugs & crashes and cheaters always count as complaints.
- Results are re-weighted to match the game's real score, so "23%" means about 23% of all its reviews.
Performance
performance, fps, frame rate, framerate…
Bugs & crashes
bug, bugs, buggy, glitch…
Price & monetization
price, priced, pricing, expensive…
Servers & matchmaking
server, servers, matchmaking, queue…
Cheaters
cheater, cheaters, cheating, cheat…
Community
community, toxic, toxicity, teammates…
Co-op & multiplayer
co-op, coop, co op, multiplayer…
Story & characters
story, storyline, plot, characters…
Gameplay & combat
gameplay, combat, mechanics, gunplay…
Graphics & art
graphics, visuals, visual, visually…
Music & sound
music, soundtrack, ost, sound design…
Content & grind
content, endgame, end game, replayability…
Difficulty
difficulty, difficult, challenging, challenge…
Developers & updates
dev, devs, developer, developers…
Accounts & launchers
account, accounts, psn, launcher…
Judging tone
- A TF-IDF and logistic regression model, trained on about 426,000 reviews from 100 games with each review's thumbs up or down as its label.
- Tested on three games it never saw (Helldivers 2, Cities: Skylines II, Hollow Knight: Silksong), it was right about 91% of the time.
Flagged events
- "Usual" is the median of the previous 6 months.
- Negative surge (a review bomb): negative reviews at least 2× usual and 300 more, with the score at least 10 points below usual.
- Score drop: 500+ reviews and the score at least 20 points below usual, without that surge.
Known limits
- Keywords can't read context: Helldivers 2's enemy "bugs" count as bug complaints, and Stardew Valley's "money" is in-game gold.
- Sarcasm and negation ("not bad at all") can fool the model.
- The text sample leans toward helpful reviews, so scores come from Steam's totals.
- A flag shows something happened, not why, or whether it was fair.