Connecting the Dots

Illustration by Jillian Tamaki

It’s easy to complain about lazy music critics who define the latest act with a recipe. For example: Norah Jones + Leslie Feist = Keren Ann. Or Serge Gainsbourg + Le Tigre = Stereo Total. Such musical equations are convenient shorthand for critics who are paid to come up with actual adjectives and original criticisms. But this new mode of definition-by-association is growing at a clip, and it’s symptomatic of how we now learn about new music, books and movies.

Describing or categorizing any new cultural product is taxing and time-consuming. There are more media for us to choose from than ever before, and much of it resists easy definition — just try explaining the sound of hot young London singer M.I.A. without falling back on a recipe of influences. It’s no wonder then that the next evolutionary step in a continuum that began with pop culture algebra is “recommender applications.”

These web-based applications seek to recommend music not through descriptive reviews, but through affinities calculated by a computer algorithm. Websites like Movielens.org and Filmaffinity.com endeavour to tell you what you will like by having you tell them what you already like. This form of definition-by-association is supplanting the good old-fashioned review as the primary way for consumers to discover new music, movies and literature.

Using a recommender application is like consulting your friends for music or movie advice, except on a grander scale. Defining by association involves abandoning micro-judgments (e.g. one critic’s opinion, your brother’s testimonial) for the world of global opinion — computer-produced taste groups based on mutual affinities. We’ve long been in the habit of defining things in the context of other things — so and so writes like so and so, someone sings like someone else. Sites like Musicplasma.com, Music-map.com, Filmaffinity.com and Movielens.org just take it to the next level, dispensing with the articles and adjectives.

Amazon.com has made casual product recommendations for years. Buy one book and you get the occasionally helpful — if mysteriously calculated — message, “Customers who bought this book also bought,” which rhymes off a list of books the online retailer thinks you will like. Not surprisingly, when I plugged in the title of a book I recently read and enjoyed, I found that four of the five books recommended were ones I was interested in, or had already read. But what if the information Amazon had about me went beyond the handful of books or CDs I’ve purchased from them in the past few years? What if they could draw upon vast amounts of data about my tastes? This is the crux of the recommender application – the more info you give, the more appropriate the recommendations you’ll receive.

Movielens.org is a research project run by the University of Minnesota’s computer science program. Over the past few years, they’ve culled more than eight million movie ratings from more than 80,000 members. Using a scoring system that runs from zero to five stars, users can rate as many films as they like. Users then receive a list of recommendations from Movielens’s extensive database. I spent about 10 minutes rating 70 movies, and then received a list of about 20 movies the program was sure I’d like, from A Very Long Engagement to a handful of vintage noir films — all of them movies I'd be interested in watching.

“Collaborative filtering is a technique for personalizing data, where we use the opinions of a whole bunch of people in some large community to come up with an individual estimate of how much people will like a certain thing,” says Professor Joseph Konstan, co-director of the GroupLens research group, which operates Movielens. In other words, by studying the preferences of vast groups, Konstan’s researchers can make recommendations about the tastes of individuals who reside in those groups.

What need have we for a “two thumbs up” recommendation when we can plug in a handful of favourite movies and receive a pile of new titles we’re bound to like? Pablo Kurt, the creator of Filmaffinity.com — a movie-recommending site with more than 27,000 members — says the benefit of such networks of affinity is that they broaden one’s tastes. “Some people have never watched a single movie by Fellini or David Lynch, or even great independent movies, because nobody told them they were awesome,” says Kurt, via email from his home in Madrid. But by sharing picks with Filmaffinity, users get a broad array of recommendations from an international database of users.

Musicplasma.com and Music-map.com let users plug in artists they like, but instead of displaying results visually, they show them as family trees. Your chosen artist sits in the middle of the page, while bubbles with other artists’ names are positioned at varying distances that are intended to reveal how closely they relate, musically, to your pick. MusicPlasma takes some of its data from Amazon.com, and so provides an enhanced representation of the information you’d get if you searched for the same artist on Amazon. MusicPlasma’s creator, Frédéric Vavrille, wanted to move beyond linear data.

“A list is an oriented, one-dimensional approach,” says the Paris-based Vavrille. “We are used to seeing lists everywhere on the net, but it doesn’t mean that’s the best way.”

Music-map creates a much larger grid than MusicPlasma and offers a much broader range of artists, providing more potentially accurate choices. For example, I typed in the name of alt-cabaret darling Rufus Wainwright. The artist MusicPlasma determined was closest to my search was rock band Son Volt. I was surprised, given that I can't imagine any rock journalist saying, "If you like Rufus Wainwright, you'll love Son Volt." I like neo-alt country well enough, but it doesn't have much to do with Rufus Wainwright or his lilting, neo-baroque sensibility. On Music-map, however, Rufus was surrounded by mostly piano-playing male musicians, which is an obvious, but also much more sensible result. (For some reason, middling R&B siren Deborah Cox is separated from Rufus by only a few pixels, proof that computers still have a way to go in working out our logical affinities.)

Ultimately, recommender applications don’t abrogate the need for written description — they just modify it. “I would not say that we’ve reached a stage where for most things we don’t have need for descriptive content,” says Professor Konstan. “[People] use Movielens, then they look those movies up. They consider 10 [titles], do further research, and say, ‘These are the two that I’m really interested in [renting].”

The looming question is whether computerized associatives (geek speak for recommender applications) will ultimately produce vast grids of preference groups – imagine a world where data mining could tell us that people who like havarti cheese and the Killers are almost guaranteed to like the colour pink. Konstan doesn’t think we’re in danger of such infinitesimal sociological grouping.

“Getting agreement across multiple domains is much harder. For example, if you went into a large football stadium, and said, ‘How many people have similar taste to me in music,’ many would. But then if you tried movies, it might fall apart. People have tried to go across domains and sometimes it works, and sometimes it doesn’t work.” It makes intuitive sense. Though you and your cousin might share identical taste in music, his favourite movies might still be your least favourite.

But this novel, tech-enhanced way of discovering new CDs and movies seems no less accurate than our now slightly outmoded means of assessment: actual reviews. Perhaps we’re heading in this direction because we haven’t got the time to read an article about a band we’ve never heard of. We’d prefer an immediate comparison on which to base a yay or a nay. Indeed, there’s a certain streamlined beauty in the concise way a record company attempts to cast Skye Sweetnam as the new Avril Lavigne – you don’t need to waste time deliberating over the former if you already have an opinion on the latter. And who doesn’t have an opinion about Avril?

Sarah Lazarovic is a Toronto writer and illustrator.