One of the more common adjectives some observers use to describe the results one gets from StumbleUpon is “random.” So what does “random” really mean?
The term “random” in the English language dates back to 1561. It was borrowed from the Old French “randir” which has roots in the Old High Germanic verb “rinnan,” meaning “to run.” In modern English, “random” refers to having no specific pattern or purpose, lacking any definite plan or prearranged order. But is StumbleUpon really random? Or is there something more to it, a sophistication that is not immediately obvious to the untrained eye? Let’s take a look.
Is StumbleUpon Random?
StumbleUpon was built on a simple concept: click a button, and be transported to a page on the Internet that you’ll love. But how does StumbleUpon choose what specific page to show you out of the billions of URLs on the web, with seemingly so little information to go on beyond some basic demographic data and self-selected broad interests? Consider that YOU are unique, have very particular tastes and interests, and that StumbleUpon aims to find the right content at the right moment for each of more than 15 million different individuals. Frankly, a “random web-page generator,” as some have described StumbleUpon, would be relatively easy to build, and would probably hold the attention of a decent number of folks for some period of time. But where’s the fun in that? There’s a reason we’ve assembled a really smart team of technologists, scientists and business people to address the universal (and increasingly acute) problem of information overload by re-inventing how people discover content and information they’ll love.
Explore and Exploit
We employ two major strategies to determine what content to recommend when you click the “Stumble!” button:
“Exploit” – Using this method, we suggest content based on your “preference model,” which we create and continuously update based on a myriad of signals you provide to us. For example, if you tend to like more videos than photos, we’ll recommend more videos. If you enjoy more content from a specific domain (e.g., Flickr or National Geographic), we’ll recommend more from that domain. If you tend to prefer pages that have also been liked by a particular Stumbler, we’ll recommend more favorites from that person.
“Explore”- Using this method, we suggest content that is beyond what we confidently know to be your preferences: Content we think you might like, but we don’t know for sure. But even these recommendations are determined probabilistically, so they’re far from “random.” We think these exploratory stumbles – often described as serendipitous or unexpectedly relevant – make StumbleUpon the best at introducing people to content and information they would never have known to look for.
Each Stumbler has a unique combination of “explore” and “exploit” ingredients in his or her stumbling recipe. For some, there’s more “explore,” while for others there’s more “exploit.” Whatever floats your discovery boat! But none of it is random!
Social vs. Implicit Graph
Whether exploring or exploiting your preferences, we source content from both social and peer graphs. Your social graph consists of people you know in real life, like your Facebook friends and your email contacts, with whom you’ve chosen to connect on StumbleUpon. While your social graph is great for communications and staying in touch, you may or may not share common interests and often you won’t like all the things your friends and acquaintances like.
Your social graph is a good starting point from which we can begin to learn your preferences, but a content recommendation system built entirely around your social graph wouldn’t produce terribly compelling results. Given the limitations of one’s social graph, StumbleUpon is unique among search and discovery services in that we also heavily weigh in one’s interest graph. Your interest graph consists of people with similar tastes as you, but whom you may never have met – and in most cases never will (though we’re proud to have facilitated several real-world matches).
The people in your interest graph are your discovery soul-mates; they enjoy the same kind of oddball humor, funky fashion sense or quirky design aesthetic as you. StumbleUpon implicitly connects you with these like-minded users based on similar “interest DNA.”
We trust you’ve got the point that behind every deceptively simple click of the “Stumble!” button is an incredibly sophisticated and computationally intensive scramble to process billions of pieces of data in real-time to produce a single recommendation. So next time you’re describing StumbleUpon to your friends and family and feel the urge to use the word “random,” remember this post and let them know instead that StumbleUpon is a highly personalized discovery engine that helps you find the best of the web. They’ll thank you for it.