Recently I participated in recommender system contest, where I managed to place among the top 10. In this notebook I provide the code and comparisons of my Item Based K Nearest Neighbours with some other algorithms.
After my best system was a plain ItemKNN. On the competition platform I achieved 38.3% accuracy calculating recommendations on the whole dataset. This is a quite good result. It could be better, if I had an inside about time factor of listens, meaning a time sequence of listens. Further improvements can be achieved being provided with more information about user.