If you look at the embedded documentation you’ll see that it talks about 3 other models, however there isn’t enough time to build all of them during this Lab. They involve the use of additional data files - a user demographic file, and an item metadata file, both of which are supplied with the Movie Lens data set in your SageMaker Notebook.
Because they required additional datasets, you need to create each of these within their own Personalize Dataset Group, and you also need to re-import the original interactions file DEMO-movie-lens-100k.csv that you uploaded into S3 during the notebook - this is because Personalize trains solutions on all data files within the Dataset Group.
The three models that you should build are as follows:
Observations are that demographics are absolutely not a good indicator for movies recommendations, nor for things like book recommendations - this isn’t an issue with Amazon Personalize, rather it is a know issue with using age and gender to predict likes and dislikes of media.
Also, the single, compound genre certainly seems more accurate for the first 5 or 10 responses, but for the set of 25 response as a whole the multiple genre model probably gets a better list of movies than the compound one.