14. Recommender Systems
Introduction
We encounter recommender systems in the world around us on a daily, or near-daily basis. They are an essential feature of most content streaming apps (music, movies, tutorials, etc.) and of nearly all app-based or touchscreen-based food ordering systems. In theory, at least, recommender systems deliver a ‘win-win’ that benefits all of the involved parties; they steer consumers towards items that are likely to be of interest to them, and they lead to additional transactions, thereby giving a boost to businesses’ revenue.
From the consumer perspective, a major shift engendered by these systems is that they make our user experiences more personalized. Years ago, everyone ordered from one menu, or one catalog, that looked identical to everyone else’s. When one person turned on the television at 7:30 p.m. on a Tuesday evening, she saw the same content that her friends and neighbors saw. Now, though, our experiences are much more tailored to our own expressed interests and tastes. As evidence of this, go to YouTube, either through the app on your phone, or via the browser on your personal computer. Take a look at what’s presented to you on the homepage. What do you notice? Next, ask a friend to do the same thing. These differences are remarkable, and they are the result of years of experimentation through trial and error from researchers working on recommender systems.
Recommendations can be either personalized or non-personalized. A non-personalized recommendation could be based on general popularity; for instance, if the Ferris Bueller ride tends to appeal broadly to Lobster Land visitors of all ages and ride preferences, then we can offer that up as a general recommendation to someone who asks us about a ride to take during a visit to the park. To make a personalized recommendation, we would need to first know either something about the targeted person (e.g. demographic data like age and gender) or something about the person’s known interests (e.g. a stated interest in tall roller coasters).
How Users Express Interests: Implicit Ratings and Explicit Ratings
Many recommender systems are built from the explicit ratings delivered by users. Consumers generate such ratings when they take conscious actions to share feedback about products and services, either directly with the service (through feedback forms and consumer surveys) or through a third-party review site, like Yelp or TripAdvisor.
You might be wondering, “Wait, how do I fit into all of this? I’ve never written a Yelp Review, and I don’t ever plan to. Write a TripAdvisor review? Yeah, right. I don’t even read those things. I don’t even ‘thumbs up’ the content that I like on YouTube.”
The quote in the paragraph above may describe you, but guess what? You still implicitly rate products through your actions as a consumer.
Think about your online actions, and even your offline purchases. Sure, you may not have reviewed them on Yelp, but why do you visit the same coffee shop down the street, morning after morning? Why do you adhere to the same Friday evening routine of ordering pizza and salads from the wonderful brick oven pizzeria around the corner from your office? Your decision to keep coming back is an implicit rating, as is your decision not to skip certain songs on Spotify, or not to abandon some YouTube video halfway through.
Implicit reviews can be negative, too. Whether it’s a restaurant you visit once but never return to, a song you always skip when it comes up on your streaming app, or a type of Facebook post that you always scroll past, without stopping to read or engage, your decisions to disregard certain options are an unspoken statement that you would rather spend your time and/or money on something else.