KNN for this dataset feels weird - with labels as 1/0 for movie genres, it feels like overkill. A rough equivalent would be just a comparison of how many tags are shared.

KNN gets a little more interesting when the features are on different scales and not binary.