Summary

In this chapter, we discussed the main techniques for building a recommender system. In a user-based scenario, we assume that we have enough pieces of information about the users to be able to cluster them, and moreover, we implicitly assume that similar users would like the same products. In this way, it's immediate to determine the neighborhood of every new user and to suggest the products positively rated by his/her peers. In a similar way, a content-based scenario is based on the clustering of products according to their peculiar features. In this case, the assumption is weaker, because it's more probable that a user who bought an item or rated it positively will do the same with similar products.

Then we introduced collaborative filtering, which is a technique based on explicit ratings, used to predict all missing values for all users and products. In the memory-based variant, we don't train a model but we try to work directly with a user-product matrix, looking for the k-nearest...