We should add an example for using Spark and MLlib to build an item recommender.
1. The data generator does not generate user product ratings. We need a way to provide a metric for the "strength" of an interaction between a user and product. This could be the normalized purchase frequency for each product. Further evaluation is needed.
2. How to evaluate the recommendations. We will want to divide the user data into 2 groups: validation and training. For the validation group, we may want to drop certain products and see if the recommender fills in those products or something similar.