Reko is a Web API to match similar MyAnimeList users and get anime recommendations. Works thanks to MyAnimeList API beta v2.
Reko matching algorithm is hash based, it uses a 64-bit hash with each position representing an anime statistic. The latter are ordered by mal popularity weighted for mean score.
complete reference, example:
0PG-13 rating,112 Episodes series,2Aired in the 10s,3Action genre, . . .63RX rating
The hash is generated from a user list, using the statistics of the last 256 anime he's watched. Each bit is assigned 1 when the corresponding stat has times watched * scores greater than the next one.
In few words, the hash roughly stores information on what the user watches and scores more than average. Similar users generate similar hashes and vice versa.
Users are compared by computing the hamming distance on their hashes.
Since genres and themes use most of the hash space (45 bits) the hamming distance is computed twice, first on the entire hash, then, with a mask that removes genres and themes bits.
The result is a number between 0 and 83, where 0 is 100% similarity and 83 is 0%.
Anime recommendations are taken from other users lists.
Recommendations from the most similar users lists are ordered by the following function:
entry scores / (entry count + 1) - floor(avg. years from entry update / 2)
Recommendations from specific users are ordered by:
entry score - floor(years from entry update / 2)
The Reko API docs are available here.