TasteForge

Find what to play next

Add the games you've played. TasteForge learns what you spend your hours on (genres, prices, play styles) and finds what fits you next.

How the recommender works

Recommendations come from EASE-safe, a collaborative-filtering model trained on the libraries of thousands of Steam players. It recommends games that players who played the same games as you went on to play, and names which of your games pushed each pick up.

EASE (Embarrassingly Shallow Autoencoder) learns how strongly each game predicts every other game. Your scores are the sum of those links from every game you've played. The safe step then re-ranks the top 50 so likely misses move down (they aren't removed): games whose tags look more like the games you quit in under 2 hours than the ones you kept playing, games most people finish in 2 hours or less, and games with negative Steam reviews.

The model only knows games that 100+ players in its training data played (about 13,000), mostly from 2017 to 2023, so very new or niche games won't be recommended yet.

Your Player Profile (genres, prices, play styles) is built from your hours. Switch to Advanced (top right) to set hours per game or cap the budget. Use Import from Steam to fill the list from your library. Everything you enter stays in this browser.

Live data · ~13,000 Steam games the EASE-safe model knows, served from Modal