osu!Oracle recommends beatmaps based on collaborative filtering and Item(Beatmap)2Vec. It creates vector embeddings for beatmap-mod combinations based on how often they are seen together, yielding a 15D vector for each beatmap-mod combination. Cosine similarity in this new vector space is representative of semantic similarity, or in this case, how similar two beatmaps are. This lets me recommend new beatmaps.
Check it out at https://osu-oracle.com
Check it out at https://osu-oracle.com