The Other End of Semantics

I wrote about Songkick last week, praising its focus on technology for the mass consumer and referenced an impending announcement. That announcement came yesterday, with the launch of a music recommendation service. Put three bands you like into the system and it returns recommendations of area concerts you might enjoy. Simple but brilliant. It works pretty nicely, too. I typed in two current obsessions – The National and Vampire Weekend – and one mainstay, PJ Harvey, and it returned Radiohead and The Cure. Two concerts I’m actually interested in seeing and will try to get tickets to. It should be said that I did stump the engine by throwing U2 in once. But perhaps it’s trying to tell me I need to update my music library.

What I like about Songkick, as previously mentioned, is that its creators aren’t interested in parsing the ins and outs of the technology. They instead want to spread their love of music through enabling technologies. I called it “music semantics” and, though the pundits in the semantic realm may take issue with that label, it’s time we embraced apps that are less wonky in their approach and focus. While Twine, Hakia, and MetaWeb are laboring in the code mines, working to build what will be the framework for the semantic Web, companies like Songkick are out in the market, showing consumers real-world applications of semantics. It’s vitally important all such players are represented, in order for semantics to develop fully and organically.



  1. stepwinder said

    Was very excited to take a look around after Carla has shared so much excitement over Songkick. I live in Austin and spend a good amount of time looking up music on Myspace. While I might not get out to see as much live music as I should, I have a reputation for always knowing where the best show is happening if a night out should go that direction.

    The potential of one site streamlining this process for me was very interesting. I’ve only played with the search function on Songkick but can’t help but wonder how far and wide the net is cast for returning events. For the first couple of searches, it seemed like Mary J. Blige, Avril Livigne, and The Cure showed up regardless of my search string. It’s possible the bands I entered were too similar in nature but I wasn’t paying close enough attention until the results felt too static. How are these recommendations made? Are there rather flat connections hardwired into the logic of the application (i.e. people who like A, might also like B, C, or D) or is the logic more dynamic (something along the lines of Pandora’s music genome projec)t?

    Two concerns….with a city as diverse as NY, the results never included bands I didn’t already know or who don’t already get plenty of mainstream radio play. I’m suspicious that this will be a tool for discovering new music. Honestly, this is the reason I didn’t dig any deeper into the site.

    And, second, when searching for a favorite local musician (Bob Schneider), results seemed to include any solo singer/songwriter with any modicum of success in Austin. It isn’t that the sound is similar. I really couldn’t figure out how the connection was any more robust than, “hey, here’s this other guy with a guitar who plays music in Austin.” It often seemed like one characteristic dominated the results. For example, when including Spoon in my search, Songkick returned the news that they were touring in NY but then only offered My Chemical Romance in the rest of the results. What about the rest of the search (Killers + Interpol)? Nothing for them? Again, the results seemed really flat.

    Those are my concerns but I have to give a big kudos too. When searching for Bob Schneider + The Hives + Black Rebel Motorcyle Club another local favorite showed up in the results, Ghostland Observatory. They’re hard to categorize (as is Bob) and they’re rarely heard anywhere other than college radio stations. This was one result seeming to point somewhere other than the usual headliners. I was excited to see Songkick provide an opportunity for others to discover Ghostland.

    I’m still interested in the concept but I’m walking away from my first “test drive” thinking there’s not much there for me just yet.

  2. Michelle said

    We’re experimenting with a very new recommendation system, so we’re still tweaking it to make it right. Your feedback is really helpful for us to get there. Concert recommendations are fundamentally different from traditional recorded music recommendations, because the set of artists playing live is different from the set of artists with recorded music released. (Think of your friend’s tiny up-and-coming band playing at your local bar’s open mic night vs. dead artists with albums out, like Frank Sinatra.)

    So in order to make concert recommendations, we have to get our set of similar/related artists in a different way. does collaborative filtering (person A has Interpol and the Killers mp3s; person B has Interpol and Joy Division mp3; therefore recommend person A Joy Division) and Pandora has experts hand-labelling music to determine similarities. We do something different. We take the entire web as a data set from which to determine whether two artists are similar. So we scan everything we can get our grubby hands on, from blogs, to Wikipedia articles, to Amazon and MySpace, etc. to see when artists are mentioned, and based on how they’re mentioned alongside other artists, we infer a similarity. This allows us to get right down into the long tail of live music, having data on every last tiny band out there.

    The first pass at this was just to make sure our machine learning algorithms are “reading” the pages right, identifying artist names correctly, making sure we can tell the difference between a music blog vs. a Wikipedia page (so we can weigh their contribution differently), etc. The next stage will be to refine the recommendations, so we can dial up or dial down the obscurity levels. Right now, it’s pretty broad brush, so if you search for a mainstream artist (like Avril Lavigne), you’ll usually get recommended mainstream artists.

    The more we learn about your taste, the better we can fine-tune the type of recommendations we give you. If you download the Songkicker (the plug-in that scans your music library) and give us access to a better representation of your music taste, the recommendations will be much better.

    I hope that sheds a little light on the technology behind the recommendations. I’m sorry your visit didn’t convince you to dig deeper, but I hope you’ll keep an eye on us as we improve and give us another chance to convince you!

  3. Shellee said

    I appreciate the detailed response and will check back later. I’m married to an IT security engineer and some of his privacy/security issues have rubbed off. I’m not wiling to give access to my music library at this point…you’re going to have to coax it out of me.

    But I promise to look again and give Songkick the opportunity.

  4. […] consumer opinion, recommendation, Semantics, Songkick I wanted to dash off a quick note and point everyone to an interesting conversation that happened here on The Guidewire. I’ve written […]

  5. Songkick looks great. If they can distinguish their service from the already strong events offering then it will be a very powerful service. I think the most exciting thing about Songkick is the thinking and theory behind what they are doing. There are obviously some very talented people in the Songkick team. Highlights such as their ‘Battle Of The Bands’ and their blog plugin ‘Bandsense’ show glimpses of their full potential and the powers of today’s emerging openmediaweb.

RSS feed for comments on this post · TrackBack URI

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: