Netflix and the Recommendation System

The Power of Artificial Intelligence for Precision Marketing

Andi Sama — CIO, Sinergi Wahana Gemilang with Saptarina.

TLDR: The application of Artificial Intelligence, implemented by Netflix for recommending the next movies to watch, based on previous viewing activities.

2020 may not be the year for many of us to expect it to be. We can not do activities as we used to be. Planned family vacations have been postponed, and physical events have been mostly converted to virtual. Worldwide regional and global face-to-face conferences & exhibitions have also been canceled and rescheduled with no clear target dates even for 2021.

For professionals, working from home has been becoming the new norm for about 8 months now. We thank God if the business activities we do can still be profitable and contribute value to stakeholders (including employees who still have jobs).

Creating a work-life balance has been becoming a real challenge during this Covid-19 pandemic situation, especially since we have spent most of our time at home. One of the “me time” could be enjoyed by watching streaming movies, provided by a global streaming service: Netflix.

Netflix shows the list of current unfinished movies (“Continue Watching for SmartTV”), a suggestion based on the user’s preferred and previous activities (“Trending Now”), and a sample category in a suggested category (“Korean TV Shows”). As of October 31, 2020.

We can watch the streaming movies from the Internet-Enabled Smartphone, SmartTV, or just from a laptop’s browser. Netflix provides a seamless multi-channel (multi-devices) user experience. We can just pause a movie while watching on a laptop or SmartTV at any time, for example, then resume on a Smartphone later. For selected movies, 4K Ultra-HD quality is available.

The Intelligence on Recommending Next Movies to watch

Netflix keeps track and monitors our viewing activities to give better movie recommendations in the future. It has a history like what movies we like or don’t like, the rating we give to the movies, which movies we have finished watching and not watched for some time, or when we are adding or removing a movie to/from our list.

Viewing activity as reported by Netflix, for profile name (user name): SmartTV.

When watching for some time, Netflix can predict and recommend what kind of movies we are likely to watch in the future. The more we watch, the better the recommendation. This is one of the applications of Artificial Intelligence (AI), a recommendation system.

There is usually one trained model for each task that the system is trained on in an AI-based system. A task can be to classify the top-5 movie categories that we used to watch in a given time period. In the case of Netflix, there are mostly multiple models that are running daily for multiple predictions.

Given four data fields such as user-ids, movie titles, movie categories, and what movie categories the users like, a model can be trained by a supervised-learning approach, for example, to classify which certain user likes what kind of movie categories. The Supervised-learning approach requires that we have a pair set of data (input, output) as a base for the model we want to generate. Input and output can be in multiple fields.

Netflix seems to be running a batch process every 24 hours to update its movie recommendations for all users. With almost 4,000 movie titles at the end of 2019 (Mansoor Iqbal, 2020) and over 193 million users worldwide in April 2020 (Wikipedia, 2020), that is a huge data to process daily.

Tips: You can change how Netflix recommends the next movies to watch by removing unwanted viewing activities. Applicable for each registered user.

Sometimes, we share one Netflix account with multiple users in the family. One Netflix account can contain several users, depending on the type of plan we are being registered on.

Different people have different interests in movie categories, and the recommendation system may be recommending too many movie categories that may not be needed anymore. Luckily, we can change how Netflix recommends the next movies to watch by removing unwanted viewing activities. This applies to each user. The changes will be in effect in the next 24-hours.

Netflix shows a list of suggestions based on the user’s preferred and previous updated activities (“Trending Now”). No more (“Korean TV Shows”) as was suggested before on the top category list. A new category also emerges “US movies.” Captured as of November 4, 2020.

The Power of AI for Precision Marketing

When we start learning AI, especially in computer vision, we usually start by learning to work on image classification, then expand later to others such as image segmentation and image captioning. More advanced ones may include processing sound, a video stream, face recognition, or action recognition. Mostly, the datasets are required to train the model with the supervised learning approach.

As the discussion in this article is general, AI may refer to either machine learning or the more recent one: deep learning.

A non-visual dataset can include numbers & texts. In Netflix's case, metadata (data that represents other data) may represent Netflix's account, user name with its unique user identification, movie preferences, movie category preferences, including details on viewing activities.

Collaborative filtering and content-based filtering are two common implementations in a recommendation system, as implemented in Netflix.

Netflix shows more category suggestions based on the user’s preferred and previous updated activities (“Time Travel TV Shows”). A new category also emerges based on our preferences in certain movies, “Because you liked Wu Assasins.” Captured as of November 5, 2020.

Content-based filtering can be straightforward. Netflix recommends the next movies to watch based on the user’s viewing activity. When several users watched movies together in a certain category (i.e., action genre), similar next-movies-to-watch can be suggested to multiple users within this category. This is an illustration of collaborative filtering.

By analyzing users’ preferences from time to time, Netflix can target each user with a recommendation that is precisely targeted for that specific user’s preference. The power of Precision Marketing is to keep users engaged with the service, thus keeping the revenue flows for the business.

Netflix may suggest a completely different new category of movies (a different genre or a created new category like “Popular in Netflix,” “Asian Movies & TV,” “because you added this film to your list,” “because you liked this movie,” “because you watched this movie,” and the like).

It’s like exercising to offer something different from what the users used to watch and then analyze whether users like it. This can be used later for making future recommendations.

References

• Andi Sama, 2020, “Face Liveness Detection through Blinking Eyes.”
• Andi Sama et al., 2020a, “Image Preprocessing with Open Source Tools: A Brief Look into Image Preprocessing for doing AI in Computer Vision.”
• Andi Sama et al., 2020b, “A brief look into Data Wrangling: An Essential Supporting Skill for doing Analytics in Data Science.”
• Andi Sama et al., 2019, “Think like a Data Scientist.”
• Mansoor Iqbal, 2020, “Netflix Revenue and Usage Statistics (2020).”
• Wikipedia, 2020, ”Netflix.”