Supervised Machine Learning with Python

Limitations of content-based systems

There are several notable limitations to content-based systems that make them less than ideal in most scenarios. The first of these is the manual nature of the feature engineering, which can be extraordinarily tough given that the difficulty of collecting the data about the items can be really time-consuming, and many times, the data we're presented about an item is limited to a text description. So, we're not given this nice encoded matrix and that means we have to extract the attributes from descriptions, which can be challenging and extremely time-intensive.

Next, we end up with the largely dummy-encoded set of content vectors, meaning it's heavily zero inflated. So, naturally, our similarity computations ...