Recommendation System is a subclass of machine learning which generally deals with ranking or rating products/users. Loosely defined, a recommender system is a system which predicts rating a users might give to a specific item. These predictions will then be ranked and returned back to the user. It is often seen as a "black box", the model created by these large companies are not very easily interpreted. The results which are generated are often recommendations for the user for things that they need/want but are unaware that they need/want it until they have been recommended to them.
The definition of good recommendations helps evaluate the performance of the recommender you built. The quality of a recommendation system can be accessed through various tactics which measure coverage and accuracy. Accuracy is the fraction of correct recommendations out of total possible recommendations for. The method of evaluation of a recommendation is solely based on the dataset and approach used to generate the recommendation.