][1]Here is my concept map for making recommendations. I didn’t include code because (1) I tried to focus on the connections between the elements needed to create recommendations, and (2) the python code for doing this is already well outlined in the Matrix:Recommendations section of the Software Carpentry Tutorials.
To make recommendations, basically you need a set of previous ratings stored in some kind of database or dictionary, an item to be rated and someone to do the rating. Because most users will have rated only a small proportion of items, the data are sparse and storing them in a dictionary makes sense to save on storage space. Once a user has rated an item, the profile of the user’s ratings are compared to the other users in the dictionary to create a similarity index between the different users (the exact same thing can also be done for the items). From a predetermined similarity criterion, all items that pass criterion and that the user has not previously ranked can then be given as recommendations to the user.