My ears and mind received a little holiday gift this week from my colleague, the inimitable DaleM. He shared a link to Pandora, a creation of the Music Genome Project. I’m floating through work this week listening to tunes based on my selections and preferences, and I haven’t heard any repeats so far.
Personalization reigns here. It starts with the abundantly user-friendly question “Can you help me discover more music that I’ll like?” I create a “radio station” by identifying an artist or song that I like. The program generates a play list based on attributes of my first choice, with descriptors like “basic rock song structures, mild rhythmic syncopation, extensive vamping, mixed acoustic and electric instrumentation, and a vocal-centric aesthetic.”
As my station plays, I can guide the programming by choosing from four options:
- I really like this song –play more like it!
- I don’t like this song –it’s not what this station should play
- Why is this song playing?
- I want to add more kinds of music to this station
When my initial choice of Paul Simon started generating a little too much schmaltz for my taste, I could spike the list with the slightly more edgy Tom Waits or Ani DiFranco. When a song came up that I really did not like, I chose the second option and got this endearing response: “Sorry about that. We’ll never play this song again.” And yes, you can share stations with your buddies.
What does this have to do with libraries? Many have suggested that library catalogs should add Amazon-like book recommendations. I’ve been lukewarm to that idea, based on my own experience of the often inaccurate and ridiculous picks that Amazon comes up with. Their algorithm is just not smart enough to reflect the nuance and unpredictability of individual choice. But Pandora hands enough controls to the user to shape the experience, yielding a rich exploration of that long tail. I’d never heard of the Buttersprites or their song LuvLuvLuv until Pandora pulled it out of my “Paul Simon” surprise box and tickled my ears.
Wouldn’t it be cool if the library catalog could do this with books, cds/dvds, and even database articles? Patrons would plug into their customized feeds, not simply to access specific information, but to expand their horizons with the discovery of things they didn’t even know they liked and needed.
Of course, this kind of development is for the big guys, the large library systems with the 5.5 FTE Web dev staff. For all of those smaller libraries with the .05 FTE Web dev staff, if only Santa would bring us a set of easy do-it-yourself modules for creating all this cool functionality on our own sites. In the spirit of the season, maybe the large systems will act like R&D for innovative development and share what they learn with their less well-endowed community members. Ho-ho-ho?
A major part of the problem with the Amazon recommendation system is that folks use their account to buy gifts. So my insterst in tech, sci-fi and graphic novels gets mixed in with the books on cooking and fly fishing for my father-in-law. Buy enough books for enough different folks and the recommendations just become useless.
A system like Pandora, where it is only you picking out items of interest to you can be much more accurate, even if the matching software is not as good. Maybe the on-line stores should have a check box to indicate it is not something for me.
That is exactly the problem with the Amazon system. I’ve done some collection development projects that involved looking up numerous titles on a specific subject, like Islam. This skewed the suggestions to the point of uselessness that you describe.
If Amazon is doing any competitive intelligence, I expect they will soon be amending their matching software. My big wish is that libraries get to play with this kind of creative personalization, so we don’t lose all of our patrons to Google-zon.
At the risk of sounding a bit Ludd-ish, I think another very significant factor that would make a Pandora-like interface difficult to implement in the library world is the fact that music has some relatively basic identifiable features that translate well across individual pieces and genres of music. I put in “David Crosby” and got “syncopated, vocal harmonies, male singer” and it’s amazing how far that gets you. It may be that information in the broader sense is much more difficult to analyze in that way. Having a Genome-like algorithm is the key to making this truly work well. I’ve never been very convinced by Amazon’s algorithm, which seems heavily based on keywords in titles. Are there Knowledge or Literary Genome Projects out there?
Anyway, Pandora is totally cool. I am hooked.
Pandora Using Local Flash Storage
By now you’ve surely heard of Pandora, the nifty little radio station that attempts to tailor the music it plays to selected artists or songs. But what you probably didn’t know was that Pandora uses local Flash storage to cache
Very interesting …
I’ve been assured (?) that most of Pandora’s staff are musicians and definitely not in league with the RIAA. They’re working on a fix. In the meantime, I’m just enjoying listening –really!
I was introduced to Pandora earlier this month by my brother, who is not only a programer but music producer. I really like their system, even if I do not understand their jargon. I asked them if they could give me more information on their fields etc, but they claimed copy right.
The big differnce between Amazon and Pandora applies directly to the problem you are all having with Amazon. Pandora uses Musicans/programrs (they claim 2000) to actually sort the music and monitor the system. While Amazon is a pure algorithm.
To create a “Literary Genome Project” you would basically take reader advisories, made by either librarians or editors, and get IT people to work on the system. It would not be technically difficult, just a lot of leg work.
Good observation! It seems that there are pieces of the puzzle already in place in the library world. Every item in WorldCat is already tagged with some of the fields necessary to characterize a work Pandora-style, but would need the addition of many more colorful descriptors. With the already-networked army of library staff, it’s not far-fetched to imagine this “Literary Genome Project.” Like you say, it’s a lot of leg work, but also requires an organized effort motivated by something other than profit.
Indeed, to make OCLC a Pandora-Like creature would require the addition of many reader-created, colorful descriptors as well as the retention of traditional cataloging terminology. III (www.iii.com) is working on a product like this for its next Innovative ILS.
As an aside, I just found Webjunction, and I am very impressed. Some insightful blogging is happening here.
-Dan
The Shifted Librarian has been following your progress with great interest, along with other ILS providers. MaisonBisson also has some ideas about the evolutinary direction that OPACs could take.
We’re all looking forward to significant changes. Up until now, the OPAC has really been just a computerization of the old card catalog, with some nicely added functionality of hyperlinks, but at the loss of “marginalia.” It’s exciting to see the library world shifting into gear to use technology to augment and maximize its long-established social role in society.
It’s cool to know that you’re impressed with WebJunction/BlogJunction. We are trying to catch the wave along with (or slightly ahead of?) the rest of library world.