MediaUnBlog
Archive for September, 2009
Monday, September 21st, 2009
Part 6 of a multi-part series which looks like it will keep going for at least another 18 months
Part 1 | Part 2 | Part 3 | Part 4 | Part 5 | Part 5.5 | Part 6
We have been on tenderhooks since the final call for submission for the Netflix Prize in late July. It took nearly two months before we could find out which of the two finalists, separated by a hundredth of a point, would be the winner (or, even if either would officially break the 10% barrier).
But the anticipation leading up to the announcement of the Netflix Prize winner had us at MediaUnbound thinking: is it just coincidence or is it playful calculation that the announcement overlapped with the beginning of autumn? This new season also ushers in a new season of television series and hit movies soon to be archived on DVD and ready to be delivered by Netflix to your mailbox. Or perhaps the timing was to frame the announcement of an entirely new contest Netflix will offer starting in the near future. It promises to be more challenging and quite a bit shorter than the last one, with prize announcements at the 6-month and 18-month marks, according to Netflix. We will have more detailed thoughts on this new contest soon but, for now, we should congratulate the winning team BellKor’s Pragmatic Chaos. Their last-minute conglomeration of algorithms and rejiggering of technology surely was the difference between them and rivals The Ensemble–and better than efforts from “186 countries” (which is 6-9 short of all of them, depending on your criteria for how many countries exist). Both teams put up such superlative efforts at the end that we think the prize might be better split between them. Then Netflix could combine the formulae from both teams and make an even better Frankenstein! On second thought, there is high probability they will be doing this already.
While we were waiting for the winner to be announced, MediaUnbound was wondering what our pal Socrates has been up to in the interim. We were lucky enough to catch up with him recently and have a dialogue of our own:
MUI: Socrates, are you still working at netSkix?
Socrates: It has been an interesting time for me, lately. It seems that my position at netSkix became redundant with the acquisition of the black box. They no longer saw fit to employ me.
MUI: We’re sorry to hear that. If you’re interested, we are always looking for good media analysts.
Socrates: What would I have to do to apply?
MUI: Just the normal application process. We will ask you some questions.
Socrates: Like what?
MUI: Well, something along the lines of what is your philosophy on recommendation systems?
Socrates: I would have to say that places like netSkix are too often focused only on trying to tell people what they would like, with not enough mind towards steering them away from what they will certainly not like.
MUI: Interesting; tell us more.
Socrates: Well, it’s like I told Polemarchus in our dialogue “Republic.” The person most qualified to protect something is also the most qualified to steal it. Thus, the technology best qualified to know what people will like is also the best qualified to know what people will dislike.
MUI: Have you heard about this great contest being held by Netflix? I believe they are a competitor of your former employer. All you need to do is come up with a formula to predict movie preferences. We think you’d be great at it.
Socrates: Fascinating. I will have to look into this. I’ve recently been doing some work for the government, devising some working definitions for the notions of virtue and justice; but perhaps those things can wait. This sounds far more rewarding.
And so our favorite philosopher Socrates may be throwing his hat into the ring for the next iteration of the Netflix Prize. We can only hope that his insights will continue to illuminate our own understanding of recommendation technology going forward.
Part 1 | Part 2 | Part 3 | Part 4 | Part 5 | Part 5.5 | Part 6
Tags: Countdown to 10%, netflix, netskix No Comments »
Wednesday, September 16th, 2009

At MediaUnbound, we believe a robust recommendation system requires constant tuning and tweaking. You can’t be afraid to try new techniques or allow technology ideology to blind you to new avenues of development.
In order to pull off this type of development approach, your team needs to be able to quickly evaluate results using both qualitative and quantitative methods. Today, we begin a regular blog series highlighting the contributions of Matt Pakulski in his role leading MediaUnbound’s human QA and evaluation team.
Matt has discovered a new type of music recommendation system and is putting it (her?) through a comprehensive review process . . .
I have a lot of records, and also I have an 18-month-old daughter. I’ve found that putting the records she pulls off the shelf (”selects for me”) on my playlist is a great way to explore the music I surrounded myself with for the last 20 years or so, now that my rate of acquisition has slowed to just a few LPs a month.
I call it the “Frances Picks” recommendation system. It has its limitations – for example, she’s not tall enough to reach records by artists beginning with A, D-E, H-I, L, P-Q, or T-V (not to mention half of S). But it beats my previous system, which involved a random number generator and a lot of counting back from the ends of shelves, and it has a human element.

That’s right, I have so many records I need a recommendation system. Why don’t I just pick the records I want to listen to? If only it were that easy! The thing is, I can’t trust myself with these decisions. So many gems are in there – records I forgot I had, records I don’t remember acquiring, records I thought I hated – if I relied on my own picks it would be so boring. Donovan, Thinking Fellers, Donovan, Lightning Bolt, Donovan, Herb Alpert… enough already!
As part of MediaUnbound’s continuing project to examine the merits of various recommendation systems, I’m proud to say I’ve been chosen to evaluate the results of the “Frances Picks” rec engine on a semi-regular basis.
This week, I present: Gateway Records 3534, “Hustle, Bus Stop and Line Dances”.
Produced by the club “Dancing Oasis” – “located on Manhattan’s fashionable East Side” – and promising “nonstop disco music at its best”, this album features a number of huge hits of the disco era (”(Shake Shake Shake) Shake Your Booty”, “Getaway”, “You Should Be Dancing”) performed by an uncredited, but energetic and tight, studio group. The focus is on the dancing – detailed instructions are printed on the back cover for a number of popular moves (a sample: “Chicago Bus Stop. Count 1. Touch R heel FWD. Count 2. Touch R toe behind. Count 3. Step RF FWD (1/4 turn R)…”)
One imagines a madcap Three’s Company setpiece with Jack and Chrissy trying to learn the “L.A. Hustle”, knocking over goldfish bowls, trying to explain their sequined getups to Mr. Roper, and so on. In reality probably this record was played 0-1 times by its original owners (it’s in like, mint condition, a sure sign of an unplayed album) and sat out the 80s on a shelf by the hi-fi, getting to know its neighbors, no doubt avatars of other fashionable movements (Jazzercise, Richard Simmons, Jane Fonda Workout, maybe a stray Leo Sayer record or two) before being sold in bulk to the store where I likely paid a dollar or less for it sometime in the 90s. A sad story really – unloved, unwanted, just getting in the way for its whole life.
Until the Frances Picks system came along!
The music, usually the centerpiece of an LP, feels almost like an afterthought on this record – just another line item. The producer is credited, in about 8-point type, along with the studio, recording engineer, writers of the dance instructions, cover photographer and “art production”, in a tiny box in the lower corner of the back of the jacket, but the musicians and vocalists don’t warrant a single mention. I guess the studio must have supplied them.
They sure worked hard for their money! Absent the syrupy string section that normally slogs fungus-like over the post-production surface of most disco records, the crisp arrangements that made the genre so easy to dance to shine out here in stark relief. The female vocalist has a mellow, if indistinct, Diana Ross-like croon, while the male vocalist isn’t afraid to get a little dirty on “Lowdown”. The spare rhythm section (just bass, a whisper-thin guitar, and a monster drummer) holds the whole piece together. True to the album’s mission, the music never stops – the drummer starts each song as soon as the previous one finishes. I want to believe that they recorded the two sides live in one 20-minute take each, but this is just an unlikely fantasy that I have.
These versions don’t top the super hits by the original artists in terms of quality – but let’s get real here, is quality the only reason to listen to a song? “Getaway” is a pretty great song, but do I really want to throw on an Earth Wind and Fire album? (I guess now that it’s the future, I don’t need to put on those LPs, I just have to jump on the youtube or whatever. But still) I mean if I really wanted to listen to “Getaway” I could. But I don’t. It’s much more fun for me to hear it on a record like this, take the personality of the original artist away, hear it as a composition that’s open to interpretation, and groove on the human drama going down with the disco club and the recording studio and all the middle Americans buying the dream.
The track “Lowdown” on this record was one I’d never heard or heard of – turns out it’s by Boz Scaggs! Who knew? He does that “Lido” song you always hear in the drugstore, but I didn’t know he had other songs too! It’s pretty good too, the funkiest track on the record (of course, that’s a pretty low bar). I find it interesting that since the producers seem to have tried to make this record about the dancing and not the music, they picked the least controversial songs they could think of. But thirty years on, with at least three or four pop culture disco revivals under our belt, it’s interesting to look at this document as a reflection of what some people at the core of the movement thought of as the mainstream. KC and the Sunshine Band, the Bee Gees and Tavares (”Heaven Must Be Missing an Angel” from the “Saturday Night Fever” soundtrack) are all represented, but so are a few songs that didn’t really make it into the Disco Canon: the Boz Scaggs song, Natalie Cole’s “Sophisticated Lady” and Lou Rawls’ “You’ll Never Find Another Love Like Mine”. A record like this gives you a snapshot of the era as it presented itself, distorted perhaps by the individual inclinations of the producers, but undistorted by the multiple lenses of successive decades; a picture of a living, breathing, powerful movement unaware of the doom that would soon be visited on it. For me, that’s as much a reason to pay attention to a record as quality.
I JUST REALIZED! “The Hustle” isn’t on the record!
So how did the Frances Picks engine (henceforth to be known as the FPE) perform in this case? How do I rate my recommendation? If my criteria are quality of listening experience and likelihood that I would have picked the record on my own, it’s an unqualified success. Never in a hundred years would I have selected this record on my own, and I thoroughly enjoyed the playing. I played it three times! I did not try to learn any of the dances though – I think that’s probably for suckers. My daughter has taught me that wiggling and writhing are the best dances. Or, maybe that’s what I’m trying to teach her.
Tags: boz scaggs, bus stop, dance, discovery applications, hustle, matt, music, music recommendation, recommendation apps, The FPE, vinyl No Comments »
Thursday, September 10th, 2009
Another installment of “What did we just chat about for 3 hours?!” taken (with minimal editing) straight from the MediaUnbound chat server logs . . .
[11:28 AM]<fingal> I think I don’t really like magnolia electric company that much
[11:28 AM]<michael> they are a really tight band
[11:29 AM]<michael> but def more southern rock than songs: ohia
[11:29 AM]<matt> I heard that on the college radio last week
[11:29 AM]<michael> yeah they are on the college radios
[11:29 AM]<matt> it didn’t sound as southern rock as I remembered them being live
[11:29 AM]<michael> the title track is southern rock like neil young
[11:29 AM]<matt> it did sound like neil young
[11:30 AM]<matt> just, not as much as I recall
[11:47 AM]<royjefflee> y’all need to get your southern rock sorted out from your country rock and folk rock
[11:47 AM]<royjefflee> Neil Young is not, and has never been southern rock
[11:48 AM]<royjefflee> cause, like, “a Southern man don’t need him ’round, anyhow”
[11:48 AM]<royjefflee> and MEC is neo country-indie rock
[12:05 PM]<matt> I guess I meant “hippy music”
[12:05 PM]<matt> “boring-ass hippy music”
[12:06 PM]<matt> which is broad enough to include neil, the allmans, and CCR, who I actually thought magnolia sounded just like when I saw them play
[12:08 PM]<royjefflee> yeah, none of that is Southern rock
[12:08 PM]<royjefflee> which is even worse than hippie music
[12:10 PM]<royjefflee> Matt, you’re going to have to explain what is “faux” about CCR
[12:10 PM]<matt> they’re from san francisco
[12:10 PM]<royjefflee> and so . . .
[12:11 PM]<royjefflee> there are no “jungles” to “run through”?
[12:11 PM]<matt> the other day I was in this cafe and they were playing this record that sounded like Jefferson Airplane and Grateful Dead, and I asked the hipster dude behind the counter, who didn’t look like he’d be caught dead within a mile of a jefferson airplane record, what it was, and he haughtily informed me that it was the Pretty Things
[12:12 PM]<matt> cause like, I guess they’re hipster-approved
[12:12 PM]<erich> did you tell him “this sounds just like the dead, dude!”
[12:14 PM]<matt> I was gonna tell the dude it sounded like jefferson airplane, but, I didn’t want to be “that guy”
[12:15 PM]<royjefflee> Matt, I really think you should always strive to be “that guy”
[12:15 PM]<matt> well like
[12:16 PM]<matt> I hated that guy when he would come into my record store
[12:16 PM]<matt> like, oh hey, this hipster-approved thing you’re listening to sounds just like this mainstream backwash that I probably don’t know anything deeper than
[12:16 PM]<royjefflee> but, he probably set you straight a few times, right?
[12:17 PM]<matt> yeah I mean that guy set me straight, but only cause I’m the type of person that’s interested in listening
[12:17 PM]<matt> whereas if you’re not gonna do any good, you’re just adding to the problem
[12:18 PM]<royjefflee> mmm — I guess I didn’t realize there was a problem
[12:18 PM]<matt> no problem
[12:19 PM]<royjefflee> I guess hipster closed-mindedness is the problem we’re really discussing
[12:19 PM]<royjefflee> and I can see how being prickly like that would drive someone further into their defenses
[12:19 PM]<matt> yeah for real
[12:20 PM]<matt> who knows if that dude was really closed
[12:20 PM]<matt> maybe he woulda been all “yeah it sounds like airplane circa bathing at baxters”
[12:21 PM]<royjefflee> I’ll never forget the day that I realized my most recent purchase — the latest (and final) Royal Trux release, sounded like the Dead, with even worse vocals
[12:21 PM]<matt> oh man I thought accelerator sounded like the dead
[12:21 PM]<royjefflee> WHAT?!!!
[12:22 PM]<matt> I remember making that comparison in like 98
[12:22 PM]<matt> it’s all good dude, I like the record
[12:22 PM]<royjefflee> you are out of your freking mind
[12:22 PM]<royjefflee> yeah, I know you like it
[12:22 PM]<royjefflee> but I will never ever fathom what goes on in your mind
[12:23 PM]<matt> I really like that song “That’s It for the Other One”
[12:23 PM]<matt> maybe I’ll play that shit
[12:23 PM]<matt> it sounds good in the summer on a hot day
[12:24 PM]<matt> and pre-famous child Courtney Love is on the back cover
[12:25 PM]<matt> oh, no she’s not
[12:26 PM]<matt> she’s on aoxomoxoa
[12:26 PM]<royjefflee> dude
[12:26 PM]<matt> anthem of the sun just has hairy dudes
[12:26 PM]<royjefflee> put down the doobie
[12:26 PM]<royjefflee> you’re going to put on the Dead because they sound good on a hot summer afternoon?
[12:27 PM]<matt> caution: do not stop on tracks
[12:27 PM]<matt> yes, that’s sort of true
[12:27 PM]<matt> except
[12:27 PM]<matt> I don’t think “they” meaning just any old boring-ass hippy music sounds good on a hot summer day
[12:28 PM]<matt> but “That’s It for the Other One” sure does
[12:28 PM]<matt> you know it brah
[12:29 PM]<matt> but I’m not playing Accelerator
[12:29 PM]<matt> nor do I think it sounds good on a hot summer’s afternoon
[12:29 PM]<matt> at least not as good as that dude crooning “the heat came ’round and busted me, for smiling on a cloudy day …”
[12:32 PM]<royjefflee> actually, the next Trux record *is* great for a hot summer day
[12:32 PM]<royjefflee> kicks off with the lyrics “I wanna go to the water park, the water’s cold but the sun is hot . . .”
[12:33 PM]<royjefflee> the difference between the Trux and the Dead is that the Dead has a victim mentality
[12:33 PM]<royjefflee> persecuted by the Man *and* the weather
[12:35 PM]<royjefflee> when I start making pronouncements like that, it’s agood sign that I should get my blood sugar levels back up
[12:35 PM]<royjefflee> time for lunch
We now return you to your regularly scheduled programming . . .
Tags: chat, godot, hispters, humans, the dead 1 Comment »
Wednesday, September 2nd, 2009
Since we are obsessed with trying to make people happy, sad, pumped up, or _____ (mad lib your mood here) over here at MediaUnbound (after all, what are good music & video recommendations but means to an emotional end?), we were especially interested in a couple of recent news items about changing measures of success in this realm.
As much as we love it, technology usually comes with an expiration date. Almost any innovation that can improve our lives is also at risk of being made irrelevant, obsolete, and eventually obscure.
Taking a broader view, one can include abstract concepts as part of our technological system. And with the recent economic downturn, there’s been a challenge to the idea of the Gross Domestic Product.
In general, the replacement of natural-capital services (like sun-drying clothes, or the propagation of fish, or flood control and water purification) with built-capital services (like those from a clothes dryer, or an industrial fish farm, or from levees, dams and treatment plants) is a bad trade — built capital is costly, doesn’t maintain itself, and in many cases provides an inferior, less-certain service. But in gross domestic product, every instance of replacement of a natural-capital service with a built-capital service shows up as a good thing, an increase in national economic activity. Is it any wonder that we now face a global crisis in the form of a pressing scarcity of natural-capital services of all kinds? [...] Because we use such a flawed measure of economic well-being, it’s foolish to pursue policies whose primary purpose is to raise it.
If this is true, if the GDP is becoming obsolete, what are our other options? Perhaps one answer is being modeled by the country currently playing the role of “national growth standard start-up” — the Himalayan country of Bhutan. As a tiny, underdeveloped country that is attempting to slowly open itself up to the rest of the world, they have rejected the GDP standard, and instead use a measurement called “Gross National Happiness.”

To create this, they’ve followed some familiar steps to get a project off the ground. They’ve come up with their own method of calculation — in this case, they judge happiness based on four “pillars,” nine “domains,” and seventy-two “indicators.” And then, they have to collect data on these categories. Without a direct link up to every Bhutanese brain, they have to resort to the old bugbear of asking people through a direct questionnaire. (A link to a pdf of a paper describing this process, complete with regression descriptions and Bentham name-drops, can be found here.)
This is obviously a tricky proposition. We’ve found that it is difficult to get individuals to rate their own experience in a music store, in a video playlist environment, and even just telling us which artists they love. Imagine trying to gather accurate data on someone’s whole life! The researchers in Bhutan have created formulas from scratch to deal with these happiness ratings, and then applying them as a constant to the entire population of 700,000 (a small number, but higher than the number of MP3Tunes users – we kid because we love you Mike).
But Bhutan is dealing with two factors. It’s population has been slowly opening up to global culture(s), after resisting it for a long time. The country only got television in 1999. In a decade, it’s gone from a land where the King was the most highly venerated personage by young people, to one where that same segment of the population venerates the likes of David Beckham and 50 Cent. Traditional values and traditional measures of happiness are surely being altered, if not lost altogether. However, things like life expectancy are on the rise. By creating a way to value happiness on a national scale, perhaps Bhutan can hope to modernize in a way that maintains the distinct cultural indicators and values that formerly were protected through closed borders. It might be working — in a recent University of Leicester study of the world’s happiest countries, Bhutan was in the top 10, despite having a per-capita income of under $2000 a year. As we’ve found, it’s all about the context and the goals.
So here’s to the incubators, to coming up with methods of crunching data that show (or prove) preferences and happiness. If other countries want start using GNH to judge their own well-being, hopefully Bhutan can start licensing the idea. It’s worked for us here at MediaUnbound so far. We hope that if they want to get this idea out, though, they will make good choices about who to help them along the way.
Tags: 50 cent, beckham, bhutan, GDP, GNH, happiness, survey Comments Off
By combining the numerical power of computers with knowledge from teams of human analysts,
MediaUnbound helps people find, discover and interact with large catalogs of entertainment
content to deliver an exciting entertainment experience. Every day people receive music,
video, concert and image recommendations generated by MediaUnbound through customers
such as eMusic, Ericsson, Napster, MTVN / Viacom, Terra Networks, NTT DoCoMo, HMV,
and TransWorld Entertainment.
The Social Web Community 2.0 Network marketing gurus all agree: Every corporation needs a corporate blog. Ours gives an inside peak into the people, opinions and activities at MediaUnbound.
What to expect: our thoughts on media recommendation technology; occasional customer and product announcements; in-depth discussions on whether MilkMoneyMaffia is best band from Greenland, or best band ever.
What not to expect: multiple posts every day; corporate babble-speak
Please direct feedback on blog posts to blargh@mediaunbound.com. For further information about MediaUnbound or other questions please see the Contact page on our main site.
|