One of the great promises of the Digital Age is a better way to figure out the answer to the question above. People love great writing, artwork, film, and music, but no one is going to experience, in their lifetime, more than a fraction of all the content in existence. That’s why we try hard to find the stuff we’ll probably enjoy.
But that’s always been really difficult – as the saying goes, you can’t judge a book by its cover. Even if you could, no one wants to waste time searching through every title ever written to find the ones they’ll like. So for ages we’ve relied on poor solutions for discovery of new content (not to mention food, fashion, software, etc.). The three main ways we’ve done this are:
Curation: Experts decide what the best content is, and we listen to them. That’s why everyone read To Kill A Mockingbird in high school, and why movie critics put out Top 10 lists. Of course, there’s much to be said for being exposed to high culture and different viewpoints, whether we want to be or not. But the nature of art is subjectivity – everyone has different interpretations and tastes, so I might not like the experts’ picks. And who decides who’s an expert anyways – have you ever bought a book from the Staff Favorites rack at a bookstore?
Popularity: TV channels, radio stations, movie theaters, and bookstores offer an array of the most popular content, and we pick from the available options. Pretty simple – they modify their offering based on what sells, and everyone wins, right? But again, there’s no personalization here, and we don’t all have statistically average tastes. Worse, picking based on popularity creates a feedback loop that might misrepresent reality (did anyone actually like Rebecca Black’s Friday video?).
Word of Mouth: The old standby. Our friends and family probably have a better idea of what we’ll like than anyone else, and we’re more inclined to trust them (I’ll read anything my dad or my buddy Tom sends me). But unfortunately their experiences probably overlap significantly with ours (as Mark Granovetter pointed out decades ago), so while you might get fewer false positives (bad recs), you’ll also have more false negatives (missed content). It’s also tedious to poll your friends every time you’re looking for a movie to watch.
Enter the recommendation engine. Of course, in the Digital Age of plentiful data, a lot of companies can get more mileage out of the same basic methods listed above – for instance, the New York Times can now easily measure and display its most popular articles. But technology can also do a much better job of helping us discover new content when our tastes take us beyond the Top Ten (this has also created a revolutionary paradigm for content sellers, which Chris Anderson of Wired termed the Long Tail). Although I’m not an expert in the field, it seems like there are at least three entirely new ways to use consumer data to recommend new content:
Intrinsic algorithms use the actual attributes of the content and combine them with individual user feedback. The best example of this is probably online radio Pandora, which uses the Music Genome Project’s 400 attributes to tag every song in its database. If you say you like a song, it cues up more songs with similar traits (e.g., beat, vocal pitch, etc.). While this approach is widely praised for helping discover good music, it’s probably harder to apply to other types of content. There are also certain things we love about great art (like a metaphor in a song’s lyrics) that can’t be reduced to digitized attributes.
Preference algorithms rely on both our own and others’ ratings. Amazon was a pioneer in using its massive scoring database to shift from just popularity-based discovery (“X is highly rated”) to adding a preference-based algorithm, too (“you liked X, and most people who like X also like Y, so we recommend Y”). But while the logic is simple, the algorithms get incredibly complex. The gold standard is Netflix, whose Cinematch recommendation engine is so critical to their success they offered a $1 million prize to researchers that could improve it by 10%. But this approach has limits too, many of which have been described by Eli Pariser (example: preference algorithms tend to be risk-averse, so restaurant recommendation engines keep sending people to decent, inoffensive places like Chipotle).
Social algorithms – right now, social networks’ role in recommendation is just word of mouth on steroids, but their use for discovery is only just beginning. You could talk about Game of Thrones (or “like” it) on Facebook today, and your friends may be intrigued. But far more powerful would be an automatically generated recommendation if a significant percentage of your closest ties liked or mentioned something.
So which approach is best? The more interesting question is how these approaches can be combined to produce exponentially better discovery. That’s why Facebook’s announcement last week that Reed Hastings, Netflix’s founder and CEO, was joining its board was exciting. Sure, it may signal Facebook’s preparations for an IPO, or its future addition of streaming video, but it might also pave the way for Netflix to integrate a social element into its recommendations. What if curation and/or intrinsic factors were added too? Google might also be well positioned to offer a killer recommendation engine in the future if Google+ takes off. And at the very least, a better recommendation system could help Amazon win the retail war against Walmart – or vice versa.
Going further, recommendation engines have been mostly add-ons for content sellers so far (stand-alone recommendation platforms haven’t been widely adopted), but imagine how powerful a universal recommendation engine across all types of content (and other choices we make) could be. Again, there are legitimate concerns about a world of excessively personalized discovery, as Pariser argues in The Filter Bubble - ideally, we’d always be quite conscious of recommendation and decide when to switch it on and off. But at the very least, I bet we’d watch a lot less bad TV.
As always, your feedback is welcome.