Have you ever gotten a recommendation for one or more videos on a topic that you were particularly interested in? Maybe you saw one video about going to the gym and now your entire Instagram feed is about fitness. Or maybe you searched for a new phone and watched a few review videos. Now, it’s like everyone is comparing that phone to all of the others. It almost seems like Instagram is reading your mind. It is not. At least, not in the way that it seems to be.
Instagram has a series of recommendation algorithms that tries to figure out what content you are most likely to watch, like, share, and so on. Something important to understand, is that your feed is not personal to you the way that you might expect. You and your friend could open up Instagram at the same time and see completely different Reels. One person might see a football post while the other sees a cooking video. Someone else might see a funny meme, while the last might see a post about computer programming. All of these people are on the same application, but are seeing different things. This happens because Instagram’s algorithm is deciding for each person what is the most interesting content for them to see based on what they do on the application.
It is constantly trying to predict what will happen next. Will you watch this video? Will you not watch this video? Will you like this post? Will you share this post? Will you watch this video again? All of these things tell Instagram something about you, so that it can continue to make predictions.
You Teach the Algorithm What to Do
When using Instagram, you might think that you’re simply browsing the application. But what you are actually doing is teaching the algorithm what to do. Let’s say you watch a 20 second video about artificial intelligence before deciding what to do. You did not like the post. You did not comment on the post. You did not share the post. But you did watch most of the video instead of immediately swiping to the next one. This tells Instagram something about you.
Now imagine doing this with ten other artificial intelligence videos in a row. Then, imagine liking one of them, and following an artificial intelligence creator. After that, imagine searching for another artificial intelligence creator. Each of these things individually might not tell Instagram much. But together, they tell a story. You are very interested in artificial intelligence. So, you want to see more of it. The algorithm learns this, and begins to show you more content related to artificial intelligence. If you continue to watch this type of content, it will reinforce this theory.
Not Only Do Likes Matter
One thing that might surprise you about recommendation algorithms, is that not only do likes matter. There are a variety of signals that can help an algorithm determine what you might be interested in. Some examples of these are how long you watch a video for, if you watch the entire video or not, if you watch it again, if you like it, if you comment on it, if you share it, if you save it, if you follow the creator, if you visit the creator’s profile, what topics you have previously engaged with, and what types of content you watch or not.
You do not have to actively engage with something in order for it to have an impact on your feed. Sometimes, simply spending more time on something is enough to affect you.
How Long You Watch Something Matters
Let’s say Instagram recommends that you watch two videos. The first one you watch for two seconds before swiping to the next one. The second one is thirty seconds long, and you watch the entire thing. All else being equal, the second video did a better job of grabbing your attention and keeping it. This is useful information for Instagram, and can be helpful in making recommendations.
It might not necessarily tell them what you want to watch, but it tells them what you did watch. It can be especially valuable if they have enough information to analyze patterns. This is where machine learning can come into play.
Where Does Machine Learning Come Into Play?
You do not need to know anything about machine learning in order to understand this basic concept, but if you want to go into more detail it is a valuable skill to have. Machine learning allows recommendation systems to analyze patterns in data.
Let’s say, for example, that Instagram knows that you often watch football highlights and analysis. It might notice that there is some pattern between what you watch. Perhaps people who watch certain football shows also like watching other people. Or people who follow a football account are also more likely to watch certain posts. Or perhaps people who watch a certain type of content are more likely to enjoy a certain type of content.
It can use this information to make predictions, and figure out what you are most likely to enjoy. It is essentially asking the question, “What is this person most likely to be interested in?”
But There’s a Problem With That
If Instagram only recommends things that you have already been interested in, then your feed would eventually become extremely boring. Let’s say that you like one football post, and Instagram recommends another one. You watch and like it, and then Instagram recommends another one. This pattern continues until your entire feed is football. This is not desirable for anyone, including Instagram.
This is why recommendation systems need to both try to engage you with content that you already know you will like, and to surprise you with new things that you might like. This is also why you occasionally see a recommendation that seems completely random, but ends up being exactly what you were looking for.
The algorithm is trying to get you to watch a football post, but it randomly recommends a cooking post that you end up liking more. You give it feedback that you like cooking, and so it starts to recommend you more cooking content, which you continue to engage with. Your feed is now less likely to be completely dominated by one type of content, but is still personalized to your interests.
What Happens If I Like Something?
Likely, it will continue to get you more content like it. But what about when you don’t like something? If you continue to swipe past a certain type of content, it will eventually realize that you do not like it and stop recommending it to you. The same can be said about other functions, like “not interested.”
Your likes and dislikes can tell the algorithm what you are and are not interested in, which is a part of personalization. This is also why your Instagram feed can be drastically different from what it was months ago. Your interests have changed, so the algorithm changed with you.
Other Companies Do Similar Things
Recommendation systems are not exclusive to Instagram. YouTube recommends videos, Netflix recommends movies and TV shows, Spotify plays music that you might like, and Amazon recommends products. Even food delivery services can recommend restaurants to you.
These are all similar concepts, in that they use what you have done to try and determine what you are likely to want to do in the future. The systems and the data that they use can be very different, but the concepts are still similar.
Is Instagram Really “Reading Your Mind”?
While you might feel like Instagram knows exactly what you want to watch, it is actually doing something much cooler. It is learning about you. If you watch a lot of running videos, follow runners, like marathon posts, search for running shoes and so on, then Instagram does not need to know that you like running.
Your behavior on the app tells it that you like running, so it can continue to recommend content about running to you. Because these systems analyze large amounts of data, they can sometimes find surprising connections and make surprising recommendations. That is why it might seem that your phone is “reading your mind.”
Why Should Students Care About Recommendation Algorithms?
If you are someone who is interested in learning about computer science, recommendation algorithms are a really interesting and accessible place to start. You do not need to be an expert in machine learning in order to fully understand them. In fact, you can create a very simple recommendation system for a project at the college level.
You could try something as simple as building a movie recommendation system, which is much easier than it might sound. You could look at what movies people have liked, and use that to determine what movies to recommend to them. This is a much simpler system than you might be used to, but it still follows similar principles to machine learning and recommendation algorithms.
There’s More To the Algorithm Than “What You Like”
This is probably one of the most important points of this article. Most people that understand recommendation systems understand that they are much more complex than simply calculating what you like. Your algorithm considers many signals, and those signals can change over time.
Your algorithm is constantly trying to predict what will happen next, and you continue to give it new information. Anything that you do on the application can reinforce or change the signals that are associated with you. Every swipe that you make, every pause that you make, every like or share or follow can all contribute to changing your algorithm.
This Is Why Your Feed Always Seems Surprising
The next time that you feel like your Instagram feed somehow knew what you were thinking about, you might be right. But it can also be something much cooler, namely that your app has been collecting information about you for the past few days. It has been learning about your interests, and it has used that information to recommend something that you ended up liking.
That is the fascinating thing about technology, and recommendation algorithms specifically. You do not always need to explicitly tell something what to do, and what you want to watch can be a lot more complex than you might realize.