AnimeCF gathers Anime recommendation lists and learns from the data to figure out similarities and differences between Anime fan preferences. Collaborative Filtering machine learning is used to find similarities and differences in Anime series, giving recommendations based off other Anime rating lists.
AnimeCF tries to build interesting inferences behind the entire dataset of ratings. This is achieved by building a set of key features that differentiates one Anime from another. Each Anime is given a set scores for each feature with positive and negative values, which measures how strongly the Anime fits those individual features. AnimeCF then figures out which of those features matter to you and gives you recommendations based on what AnimeCF thinks are your preferred features. The features are not pre-determined, they are achieved through machine learning, so the individual features that are used may not be perfectly understood. For example, one feature appears to be "Slice of Life-ness" and another seems to be "How Mecha is this?" based on the scoring it gives to each Anime, but this cannot be conclusively proven as the features are automatically generated with no name or definition. Some features are more murky, and may represent an ineffable combination of preferences. By decomposing Anime into defining characteristics, AnimeCF can give you uncannily accurate recommendations.
The accuracy of AnimeCF is high. The Root Mean Squared Error (RMSE) is below 1.25 points. Out of a point score between 1 to 10, your recommendations' score predictions should be on average no more than 1.25 points off what you will rate it after watching the Anime. Note that RMSE cannot be compared across different corpuses, other than the obvious point that these Anime ratings are out of a point score of 10 instead of 5 (like with the Netflix dataset), the data here is a lot less sparse which makes it more accurate, but there are fewer ratings which hinders accuracy as well. The only thing you should infer from an RMSE below 1.25 is that AnimeCF will probably give you Anime recommendations close to one point off what you would have rated it. This is why only listing Anime you absolutely love by giving only 10 point scores may not give you the best recommendations. It works by predicting scores, and if it only sees you giving 10s, AnimeCF will think you'll give anything a 10.
The recommendations should be relatively accurate if you have more than 20 Anime series in your list, and if you include both Anime you really like as well as Anime you strongly dislike (with low scores). It is important to give lower scores for Anime you strongly dislike and are ambivalent about, so that AnimeCF can figure out what kinds of Anime to exclude from your recommendations. The more Anime series listed on your Anime list, the more accurate the recommendations will be for you, and more accurate for others as well!
The recommendations link to Amazon affiliate links to help defray the cost of the compute cluster and website. Whether the Anime is available for sale on Amazon (or any other data from Amazon) does not in any way bias the rankings or items recommended.