Hybrid Recommenders

In this final chapter, we will discuss recommender systems in the context of practicality and industrial use. Until now, we have learned about various types of recommender, including knowledge, content, and collaborative filtering-based engines. However, when used in practice, each recommender usually suffers from one shortcoming or another.

We've discussed these shortcomings in the very first chapter (for instance, the novelty problem of content-based engines and the cold start problem of collaborative filters). We also briefly introduced the concept of the hybrid recommender: a robust system that combines various models to combat the disadvantage of one model with the advantage of another. In this chapter, we will build a simple hybrid recommender that combines the content and the collaborative filters that we've built thus far.

...

Technical requirements

You will be required to have Python installed on a system. Finally, to use the Git repository of this book, the user needs to install Git.

The code files of this chapter can be found on GitHub:
https://github.com/PacktPublishing/Hands-On-Recommendation-Systems-with-Python.

Check out the following video to see the code in action:

http://bit.ly/2uOHwcd.

Introduction

As already mentioned a couple of times, hybrid recommenders are extremely powerful, robust systems that combine various simpler models to give us predictions. There is no single way in which a hybrid model could do this; some hybrids predict using content and collaborative filtering techniques separately to produce results. Some others introduce content-based techniques into collaborative filters and vice versa.

Netflix is a very good example of a hybrid recommender. Netflix employs content-based techniques when it shows you similar movies to a movie you're watching (the MORE LIKE THIS section), as shown in the following screenshot:

Here, we can see that while watching Ratatouille, Netflix recommends movies to me that are very similar to Ratatouille. All the top five recommended movies are all animated and produced by Disney Pixar.

However, animated movies...

Case study – Building a hybrid model

In this section, let's build a content-based model that incorporates some collaborative filtering techniques into it.

Imagine that you have built a website like Netflix. Every time a user watches a movie, you want to display a list of recommendations in the side pane (like YouTube). At first glance, a content-based recommender seems appropriate for this task. This is because, if the person is currently watching something they find interesting, they will be more inclined to watch something similar to it.

Let's say our user is watching The Dark Knight. Since this is a Batman movie, our content-based recommender is likely to recommend other Batman (or superhero) movies regardless of quality. This may not always lead to the best recommendations. For instance, most people who like The Dark Knight do not rate Batman and Robin very...

Summary

With this, we come to the end of this chapter, as well as the main part of the book. In this book, we learned the following:

• We were introduced to the world of recommender systems. We defined the recommendation problem mathematically and discussed the various types of recommendation engines that exist, as well as their advantages and disadvantages.
• We then learned to perform data wrangling with the pandas library and familiarized ourselves with two of pandas, most powerful data structures: the series and the DataFrame.
• With our newly found data wrangling techniques, we proceeded to build an IMDB Top 250 clone. We then improved on this model to build a knowledge-based recommender that took into account the recommended movies' genre, duration, and year of release.
• Next, we learned how to build content-based recommenders using plot lines and subsequently more sophisticated...