Content-Based Filtering in Ecommerce Data Science

Out of the two types of recommendation systems, let’s begin with the content-based filtering. On e-commerce sites, based on your past purchases or the items that you would have looked up in the past, you are often recommended certain items. These are based on content-based filtering. Let’s learn how this is done.

Note here that although the values of item vectors are logical in this dataset, it is not always true. The values can be numeric as well as characters. Numeric values can quantify the prominence of an attribute in a particular item. For example, for an action movie (such as a James Bond film like Spectre, Casino Royale) that contains a lot more action than other action movies, the item-vector could be (9, -10, -6). The value 9 represents that it contains a lot of action stunts, -10 denotes that it is not an animated movie and -6 denotes that it is not usually liked by children.

Notice that the difference in magnitudes is now much larger than when the item vectors only contained logical values. Thus, the recommender system will be able to differentiate better between the items.