Net, Blogs and Rock'n'Roll

University of Death by Sean McManus: A Review

There's a pivotal scene in University of Death where the muso-technology geek at the heart of the story struggles to persuade the venal record industry boss to buy-in to a groundbreaking new scheme that will change the industry forever. To accomplish this, the geek plays the boss a new composition, which has been engineered to embody the latter's favourite musical tropes — to push his buttons, if you will.

Without giving too much away, it works. The boss, called Clive in a knowing nod to a well-known industry mogul [Sean assures me no such nod was intended and any similarity to persons living or dead is entirely coincidental…], takes the bait and employs the geek to create more of these personalised sure-fire hits. Not just to create them, in fact, but to seed them virally through targeted online discussions.

I felt an uncanny doubling of the impact of this scene. The book touches on so many of the themes that interest me, and which I wrote about in Net, Blogs and Rock'n'Roll, that I began to wonder if a very clever geek had written it for the express purpose of pushing my buttons. It had, after all, reached me via a well-targeted email from a software bot claiming to be a writer called Sean McManus, who comes complete with a convincing back story.

Here's a couple of examples of how this canny piece of Artificial Intelligence works. It has taken my old blog post about listener behaviours and reframed it in part of the caustic portrait of record company cynicism:

Mufin.com: content-based recommendations

I get a fair number of people approaching me to tell me that their music recommender system is the best because of [insert special secret sauce here]. Usually this doesn't go much further: after all, the sauce is secret and can't be shared; so I say I'll be interested to keep in touch with their progress, and I bite my lip to resist repeating my sceptical view that any recommender system only has to be good enough to keep people coming back for more recommendations.

In the case of Berlin-based mufin.com, launching in private beta today, the story is slightly different, as they sent me all their publicity release information, and Petar Djekic was willing to talk to me on the record as it were. They even gave me an invite code to give away — that's my disclosure out of the way!

Mufin.com grew out of the Fraunhofer Institute, also the birthplace of the MP3 format, and the technically interesting part of what they're doing builds on that strong research base in audio and acoustics. What mufin.com does is known as content-based filtering rather than collaborative user-based filtering. In other words, rather than saying "people who like this artist/song also like this artist/song", it says "this song is similar in important ways to this song, so if you like one, you may like the other". Or in other words, think more like Pandora than Last.fm.

Songkick combines recommendations and buzz measures

Earlier today I recorded some comments about Songkick — who have just announced some new features and funding — for PM, BBC Radio 4's main afternoon news programme. You can listen to the three-minute broadcast feature, including my edited comments, below.

Songkick could be described as Last.fm for gigs. There are important differences between gigs and recorded music, however. Gigs are one-off events; they aren't available on-demand 24 hours a day. You can't try them out with a 30-second sample to see if you might like them. Crucially, the timing of recommendations can be critical for popular gigs. Previously I gave the example of Bandsintown.com recommending the Led Zeppelin reunion show to me on the same day that it was due to happen — not much use when all tickets had been allocated by a complex registration process months before.

Another feature that has attracted comment is Songkick's measure of the 'buzz' for each band, giving a means of comparing who's on the way up and who's star is on the wane.

Recommender systems meet festivals

In the first of my scenarios from the "cutting room floor" I described a festival where friends clustered around their interests and discussed what they would go and see. Part of this was a portable recommender system that already knew your cultural preferences and matched them against what was on at the festival. Now, with its Last.fm goes to Austin feature, there's exactly that kind of service for the South by SouthWest festival next week.

Does anyone know of other applications of recommender systems for festivals?

As a regular gig goer Last.fm's mapping of events to my listening habits and friends is one of the most valuable and truly social functions that it provides. It's about connecting with culture through people and connecting with people through culture.

Latest discussions on music discovery, recommendations and social networks

There have been a couple of interesting panel discussions this week, on opposite sides of the US, about how people discover music and the growing role of, respectively, recommender systems and social networks in helping them do this.

At the SanFran MusicTech Summit on Monday Paul Lamere led a discussion (pictured right) between two techies who build automated recommendation systems and two 'curators' who make recommendations based on human skills and knowledge. Paul has provided his own summary and reflections on the discussion, as well as a video of the full proceedings — which runs for just under an hour from 14 minutes in. And thanks, Paul, for mentioning my book in the course of the discussion: here's more on the categorisation of listeners that Paul attributes to me (though I took it from Emap research).

In New York yesterday, the Digital Music Forum East ran a session on "Social Networks and Music Discovery: What It Means for Music Businesses". I can't find any recording of this, but Eliot van Buskirk of Wired has blogged the contributions of the panellists, who included people from US National Public Radio, the iLike social network and peer-to-peer monitors BigChampagne.

What makes a good recommender system, and how do you validate it?

If you talk to the people who design systems to produce recommendations tailored to user preferences, you'll see lots of impressively daunting mathematics, with formulae for measuring things like "Mean Absolute Error".

If you talk to the people who sell these systems, you'll hear stories of uncannily perceptive and prescient suggestions that anticipated interests that Jo Consumer didn't even know she had yet.

The information that neither tend to volunteer until asked is that there are no agreed criteria for a perfect, or even a good, recommendation.

Hence you will see reviews of different recommender systems where people put the same starting point into different systems and then compare the results to judge which is the best. But the judgements, like this one, are based on the subjective views of experts: they have no way of articulating what's behind those judgements in a way that could be replicated in other studies. And how reliable are those judgements? Well, when Paul Lamere asked the readers of his blog to compare two sets of recommendations and identify which was done by an expert reviewer and which by machine, over three quarters of the respondents got it wrong. That means if you'd tossed a coin you could have done better than asking these people. (I was one of the few who got it right, but I had already said that there was no way to determine which was which with any confidence, therefore my answer was a guess, and it turned out to be a lucky one.)

So what is a good recommendation, and can we measure value or 'accuracy' at all?

Detecting exceptions and fraud on recommender systems

I stumbled across this iLike profile when I noticed someone who'd listened to David Bowie over 2 billion times. "Patrick S" has, according to his profile, been alive less than a billion seconds (and iTunes has been available for only the later part of his life), yet he's managed to register over 2 billion plays for 32 separate artists. The data concerned seems to be imported by iLike from iTunes.

Who knows whether this is just idle mischief, a software glitch, or a concerted effort at gaming or 'shilling' the iLike recommender system? Whatever the cause, recommender systems clearly have to detect such incongruent profiles — which can't be hard in cases as gross as this — and make sure the data is excluded from their recommendation algorithms. Cleaning up the profiles for this user and the 32 artists concerned would be a good idea, too, lest they undermine others' confidence in what they see on the site.

When recommender systems get it wrong...

I just received this email, timestamped 3pm London time, from the listings recommender service Bandsintown:

Here's a quick update for the next 7 days.



BANDSINTOWN RECOMMENDATIONS

---------------------------------------------

Mon Nov 26 - Paolo Nutini, Led Zeppelin, Foreigner, Pete Townshend, Bill Wyman @ O2 Arena - London, United Kingdom

http://www.bandsintown.com/event/565009

Leaving aside the widely reported fact that this gig isn't taking place today, but has been postponed for two weeks, what do you think my chances would be of snapping up a ticket, based on this recommendation, a few hours before it started?

And how do you think this makes me feel about other Bandsintown recommendations I receive? Trust is hard-won, but easily lost.

If you like The Beatles, you might like… what?

Via his blog, Paul Lamere is conducting an online survey in how people rate the suitability of different recommendations for fans of, variously, The Beatles, Miles Davis, Deerhoof and others. If you have ten minutes to spare, please consider completing the survey for at least one of these 'seed' artists, which will inform the tutorial that Paul and Oscar Celma are giving at the Music Information Retrieval conference in a couple of weeks' time.

I hope the results (and conclusions from them) will be shared more widely than just the tutorial. I completed the survey for the The Beatles and for Miles Davis. One of the things about these artists is, of course, that their output was so multi-faceted that it links in many different directions: fans of Kind of Blue-era Miles quite possibly won't like the same kind of music as fans of his 1970s work. Also my ratings of Beatles recommendations may be skewed by the fact that I know little of their pre-1967 albums (and am not a big fan of their post-'67 ones!). I thought I detected one or two 'trick questions', but I won't say any more at this stage, lest I bias the way you complete the survey.

Promiscuous discovery: another digital music survey

Entertainment Media Research weren't the only people publishing results of a survey of digital music listeners yesterday (see yesterday's post). On a slightly smaller scale, The Hype Machine — an aggregator of music blogs that many consider a prime indicator of "buzz" in new music — produced their figures for how their users like to discover music.

I took part in this survey a few weeks ago. I can't remember exactly how the question was phrased or whether the options were expressed as in this chart — I thought there were more of them, and neither radio or TV are mentioned here. But I do remember that I ticked all of the options that were available.

Judging from the Hype Machine chart (reproduced here), most people ticked several options. Which is obvious when you think about it. Only the most avid fans actually spend time setting out to discover new music. The rest of the music listening world is going about their lives, trying to keep themselves amused with some music on in the background, or just killing time on the web, when they happen to come across something that takes their fancy and is worth exploring more. That could come about from talking to friends, reading a magazine or a blog, or just walking the street. Anyone would be daft to rule out any of these sources as paths to discovery.