| |
Pandora and the Profilers
by Jen Legier, Customer Service Bibliographer
Recently I had the great pleasure of sitting down with Tim Westergren, of Pandora.com and Rob Norton, of YBP to discuss books, music and creative classification.
How well could you describe what it is you enjoy about a great bottle of wine or that painting in the museum that you could gaze at for days? The label on the wine might toss out terms such as full-bodied, robust and fruity; the museum guide might describe that painting as an Early Renaissance work with masterful use of perspective and anatomical realism. And while you may agree with the label and the museum guide, perhaps you originally discovered your bottle of red by accident; maybe you just happened by that painting on a whim, and you most likely didn't have any of those words in mind as you browsed in the wine shop or wandered the gallery.
We don't go about everyday choices armed with a list of adjectives and if we did I'm not sure what our choices would end up looking like. So, what if it's your job to find that ideal wine or painting or car or gift or song or book, and the ideal is determined by what someone else wants? There are probably millions of descriptive words out there that could be applied to thousands of different categories, and in our digital age, they can all be accessed quickly enough, but how do you sort through all that data and come up with a gem? If you are Tim Westergren, founder of the internet radio site www.Pandora.com or Rob Norton, Senior Profiler here at YBP, you narrow down the choices by putting that massive list of words and categories to work for you.
About the same time I started working at YBP, a friend of mine turned me on to Pandora.com (check it out). The idea behind it is to start with a song or band you like and let the program (named the Music Genome Project, which would also be a great name for a band) suggest other songs that you might also enjoy. What sets Pandora apart from other sites that make recommendations is the huge amount of knowledge and unique progressions hidden behind the scene. Since Pandora's inception, Tim Westergren has hired a small staff (including a musicologist) dedicated to unraveling songs and very specifically coding them with a list of aspects and weighted scales and serious classification systems. When you listen at Pandora.com you don't get songs by the same artist or label as the one you started with, and you don't get the "other listeners who liked X also enjoyed Y" bit. This is a living, breathing, custom-tailored jukebox just for you. You can check out the list of attributes on Wikipedia, but there are deeper levels and massive amounts of intellectual energy expended to pin down the genes for a single 3 minute song. Tim and his staff appear to share a deep love for music and are committed to decoding songs consistently so their listeners find songs that match their likes and dislikes while also turning up new discoveries that they might never have heard otherwise.
As I was sitting with Rob Norton for my profiling training as a new bibliographer at YBP, I thought the ground rules sounded remarkably like the Music Genome Project. Rob took me through a day in the life of a YBP profiler and I slowly started to absorb the countless ways in which a book in hand can be described; trust me, you'd be amazed. YBP's approval customers will recognize the 3 main profile categories: press list, subject classification and non-subject parameters, and, while the first two are pretty much a yes or no, once you delve into the "NSP's" (as the non-subject parameters are referred to around here) things get a lot more challenging and a lot more interesting. (There is a list of these terms in the help section of YBP's GOBI website, www.gobi2.com). Our profilers are also dedicated to consistency in their work, and I hear their round table discussions to generate or rework definitions are lengthy and sometimes become great debates. Now that the definitions are agreed upon, a profiler who is fully trained to carefully apply them will spend the day reviewing books and making decisions about the way we describe it. This may sound like an idyllic way to spend a day, and our profilers enjoy their work, but there is much at stake. Our approval libraries have described for us what kinds of material they want, and we have translated those instructions into the rules of their very own automated approval plan which take into account the make-up of each book our profilers see; after that final keystroke, the computer matches up the customer rules to the description of the book and the library's bibliographic fate is decided (a book, a notification slip, or nothing). It's important to note here that Tim and Rob strongly agree that decisions about how to classify are not to be influenced by the opportunity to play a song or sell a book; they operate outside the realm of the subjective, and since their job does not involve selling the most records or books they can focus all their efforts on getting the labels right.
Tim's Music Genome Project might describe a song using the attributes Disco Influence, Insistent Backbeats and Subtle Use of Turntables. Rob and the rest of YBP's profiling staff may apply terms such as Advanced Academic, Devotional or Festschriften. Taken alone or considered as a group they don't mean much, but when used as ingredients in a larger categorizing recipe, a kind of alchemy transpires. I will tell you that I have purchased several CDs based on a song I heard on Pandora.com, and I know many libraries get weekly shipments of books based on their approval plan with YBP; in both cases the end user is getting results that allow him to extend the reach of his personal preferences far beyond an individual person's capabilities. There just isn't enough time in the day to search through hundreds of catalogs to find that newly released song or book that will fit your requirements so precisely, especially when you don't know that your ideal exists, or exactly what it will sound like. As a newcomer both to Pandora and to YBP, one is struck by efficiency, precision and results that each system affords.
Now, both Tim and Rob will admit that their own system of categorizing is not foolproof; there are countless outside forces to consider (copyrights, languages, bindings, song length, etc), and duds do happen, but in both cases there are built in correction mechanisms to take advantage of. On Pandora, there's a way to give your "thumbs up" or "thumbs down", which leads to an instantaneous adjustment to the program based on that particular song's makeup. In the case of YBP, approval plans are constantly being refined and re-refined, added to and subtracted from; we have even custom built a few questions about books for libraries that have extremely particular collection needs. Adaptability is a big priority for both of us, and it allows us both to keep up with our audience and the massive amount of material available and to whittle down our output until it mirrors the ideal put forth by the user as closely as possible. Undoubtedly there are scores of new ideas just waiting to hatch regarding how we handle data in the information age; Tim is currently active in the Save Net Radio campaign for fair handling of internet radio royalties, and our profilers at YBP will certainly be involved in the challenges of handling e-books as their numbers multiply. We're looking forward to the challenges and we hope you enjoy all the gems to come.
|
