29/08/2012 08:46 BST | Updated 17/10/2014 11:59 BST

Breaking the Mould: How to Build a Discovery Engine

In the beginning, there was search. With the rise of Google and its competitors in the early 2000s, search became the primary way of locating relevant web content. But this method of information retrieval has its limitations, requiring users to provide a highly specific query and have a clear idea of their wants and needs.

In the beginning, there was search. With the rise of Google and its competitors in the early 2000s, search became the primary way of locating relevant web content. But this method of information retrieval has its limitations, requiring users to provide a highly specific query and have a clear idea of their wants and needs. Today, as most of the offline world now has an online counterpart (even down to our social relationships and personal libraries of discoveries), the information retrieval challenge has become more complex. The question created by this crowded online world isn't how technology can help users find what they want - it's how it can help them discover what they didn't expect to want. In other words, how can the Internet evolve beyond fetching and towards recommending?

This is a question that many innovators are tackling today, making content discovery and curation an exciting and crowded space. The term broadly includes traditional recommendation services like Amazon, Netflix, and Pandora, but also social networks like Facebook and Twitter, search engines like Google and Bing, and vanity curation sites like Pinterest and Flipboard, which present content feeds in a new way. But while these companies are contributing great research, ideas and products, are they really delivering on the promise of showing the most relevant undiscovered content at the right time to the right users?

Here are some Do's and Don't's of online recommendations that educate, entertain, and enlighten:

DON'T Just Make the Content Look Good

Several companies are innovating in this space by displaying content in a new and engaging way. However, there's been little discussion in the media about their recommendation mechanics - how they decide what to show to certain people. What's exciting about Pinterest, for example, is that it gives users the opportunity to discover what their current friends are exploring and sharing. Flipboard consolidates users' social and content feeds into a newspaper-like interface, which is a wonderful step towards marrying old and new technology. These sites are undoubtedly fun to use and are very suited to the mobile world we're living in. But the trouble here is that these sites aren't really helping users sidestep the ordinary - they're presenting their current connections in a new light. That's valuable, but that's not going to break the mould. Recommendation sites shouldn't just build on the social and content connections that users currently have, but they should also focus on the new connections they want to make. That's how the Internet will introduce people to new things in efficient and effective ways.

DON'T Rely Too Much on Social

We can't deny that social networks have changed the Internet forever and for the better. But relying too much on social connections will keep our worlds smaller when it comes to what we read and discover. Social connections are the means but not the end for content discovery, and here's why: Your friends on Facebook, and even in real life, don't necessarily like what you like. You're connected by circumstances and experiences, but not necessarily by passions or tastes. At StumbleUpon, we don't pretend that one way of making friends is better than the other (although many couples first met through our platform, despite the fact that we aren't a dating site), but we recognise that people look to different relationships in their lives for different kinds of experiences - and different kinds of recommendations. That's why it's important to match users by their potential to be like-minded, without assuming too much from their expressed likes and dislikes or from their stated social connections. This best reflects the serendipity that can strike in real life when you discover an unexpected compatibility with a person that you never expected. In the future, we envision social to evolve to be more feature-based rather than the core of a product.

...But DON'T Be Too Technical Either

On the other end of the spectrum are sites that are too technical in the approach to content curation. Some companies rely too much on an algorithm and don't successfully integrate human curation methods. This approach stays too close to the shore, surfacing content that's contextually related to the last site, video, or photo that a user viewed or watched but that doesn't satisfy the user's appetite for the fresh and unexpected. There's an engineering term for this approach called "exploitation," where a recommendation system makes a very safe guess on what to introduce next based on a user's clearly defined preferences, such as thumb-ups and thumb-downs. But a great recommendation engine also employs "exploration" methods, probing the user with content with a high likelihood of resonating with them based on a host of other signals.

DO Bring Back Serendipity

To create high quality, relevant, and fresh recommendations, both human and algorithmic approaches are needed, and technology must rely not only on input from the user, but also on all of the implicit signals that a user provides when interacting with a piece of content. Do they share the content? How long do they linger on it? What time of day is it when they consume it? Do they like reading about the hottest trending topics, or do they prefer more evergreen, niche subjects? We've been asking dozens of these questions whenever a user arrives on a site using StumbleUpon since we came on the scene in 2002. And we'll keep coming up with more questions, because recommendation technology is a fascinatingly complicated science that's currently transforming the user's experience on the Internet.

More fundamentally, before leaders in this space can take any steps towards truly recommending new and interesting content, they must embrace the latest paradigm shift in online user behaviour as it moves beyond search to discovery, and we must acknowledge that this problem is far deeper than how content is displayed on our iPads. Because human beings are curious explorers, and we don't always know what we want (or want what we think we do!), we welcome thoughtful recommendations from trusted sources that can help us break our routines and stumble upon the unexpected. If the Internet follows its historical trend of becoming more and more like the real world, recommendation engines must satisfy the use case where users don't know what they want or what they should look for. Such conditions are ideal for setting the gears of inventive thinking in motion, leading to ideas and experiences that turn out even better than originally intended.