« Mobile Socnets + Location Not Compatible With Making Money | Main | Spanish Virtual X-ray Clinic Taps Kennet for Growth Capital »
September 13, 2007
User Profiling Gets Visual With U-lik and Imagini
If Last.fm (acquired by CBS for $280M) was last year’s taste-matching trendsetter, what will be next year’s? A couple of European sites that are onto something are U.[lik] and imagini.net. Both are taking a more visual approach to creating user profiles based on interests, tastes, and habits.
And both package the profiling - at the moment - in a social network that promise to match users based on similarities.
There is also French startup Criteo, which could probably fit in here, but we profiled it earlier.
Imagini.net is running two ‘personality’ tests at the moment. It is kind of a visual rendition of one of those personality tests that used to be staples in mainstream consumer magazines. There is one based on what you desire and another to determine a personality profile.
No registration is required to do the tests, but if you want to save the selection you have to register. About 4 million people, according to the site ticker, have completed profiles.
At u-lik, after a basic registration step, a user gets into choosing and rating based on things like stuff you already own, or things you’d like to own, see, or have and so on. It doesn’t print you up a personality analysis but it also offers matching with like minded people.
It is a bit more geeky at the moment than imagini.net. We say that not just because of its name, but because the user interface takes a few seconds longer than imagini's to internalize.
We met the founder of U.[lik], Raphaël Labbé, a couple of months ago. He’s been obsessed for the past two or three years with the idea that people are better described by what they want or what they like than what they buy. (Note that sister-site the alarm:clock will be doing a profile of imagini.net soon).
We asked him what the two sites have in common. He said that beyond a new twist on user profiling, there’s a common business model: affiliate marketing based on a personalized list of products, and selling on the aggregate data about users’ consumption habits.
Read on for a Q&A that touches on what Last.fm did right, what Amazon could be doing better, and Labbé’s view on how his technology can contribute to better online sales and online customer conversion. It is based on an edited version of several weeks of email between the a:c euro and Labbé.
What is wrong with Amazon's recommendation engine in your view?
Amazon’s recommendation engine is a very efficient sales booster, since it’s actually their second source of sales representing about a third of total sales, just after their own search engine. Considering a sales volume of $10.7B in 2006, no doubt you’d pay a tribute to Greg Linden, the engine inventor, as Jeff Bezos did.
But Amazon’s recommendation engine is not really user-customized. It’s far from being like the trusted librarian or friend who knows your tastes. That was the kind of experience we were looking for and are currently building with U.[lik].
We started U.[lik] two years ago because we weren’t satisfied with Amazon’s generated recommendations, despite being one of the most visible recommendation systems at that time, built on a large-scale platform and fuelled by a huge number of click streams.
Amazon‘s engine mainly computes selling statistics from their customers’ purchase history. It has flaws like two DVDs that end up being rated as similar because many people bought them at the same time. Flickr provides many such examples.
Also, my purchase history is not a really good proxy of my tastes: I dislike some CDs I ordered and many books I bought were intended as a gift for someone else.
Amazon has actually few social / community features that would entice users to rate items and share them. We even assume that a merchant site can’t do that business well. That could explain why Amazon is launching alternative solutions like Amapedia, or investing in a company like shelfari.
E-retailers have never created such an experience but sites like Last.fm have by providing a social and personalized experience (with a business model based on affiliate marketing and sales revenue sharing).
U.[Lik] is on a similar track, building a trusted place where users can discover entertainment and culture, rate items based on a personal scale and engage with users having similar tastes.
You wrote that the next wave of Web innovation is ‘all about math’ - what do you mean by that?
Due to the sheer wealth of information available on the web, maths and algorithms are essential to organize and filter that huge amount of data humans can’t handle.
Nowadays, the main issue is to give access to less known but relevant content. Existing algorithms are becoming inadequate because they tend to rely on popularity.
To address these problems we need new approaches and math tools that focus less on statistics and more on organizing information at a lower scale -- a few like-minded people is definitely more valuable than having a thousand contacts.
The difficulty is that computing at this micro-scale is harder because you’ve got less data, so you’ve got to be more precise.
How do you address the problem of less data and more precision?
The U.[lik] algorithm tries to mix several sources to gain in precision: computing is mainly based on the votes of users. It is completed when data is lacking with natural links between the items (the casting for movies for example) and tagging (filtering by genres for music for example).
Pandora with its Music Genome is also following this path, what we call “content based” algorithms.
Can you point to any other examples that are not based on statistics like "popularity" or "what else did I buy" ?
Pandora is unfortunately a rare example. But attempts are being made, for example with music by trying to analyse the songs directly in a project at Sun Microsystems from Paul Lamere ). I am not sure if librarything is using its huge amounts of tags in its algorithm, probably not.
You wrote that you agree with the view that we’re coming to the “end of the page view”.
Value is less in eyeballs and more in interactions and leads generation. It’s more valuable to sell “intention” like Google does when it puts related advertising to what people are looking for. Its recent move to introduce “pay per action” PPA where advertisers only pay for a result is creating another challenge for “page views” business models.
Web technologies are changing with rich internet application, Ajax and streaming. All are making pages views less relevant to capture what users really do.
Nielsen’s decision to focus on time spent to measure online engagement is a recent acknowledgment of this change.
Social networks have to deal with two issues when they are monetized by pages views advertising. Doing advertising on a user profile is much harder than advertising on a given topic.
The intention of the user is to engage socially and most social networks don’t have a clear view of what ties people together. The more the network grows, the more complicated it is to sell its inventory, especially when a real energy has been developed to optimize that same number of page views.
These two phenomena can explain why many social networks deliver lower conversion rate on ads as they grow.
Building a social network around socials objects enables to capitalize on affiliation programs, which are the first and main form of PPA.
Building a trusted context where user can opt-in to engage with brands is another possible approach. While building U.[lik] we have always kept in mind that advertising or suggestions when reaching a certain level of personalization will likely be as valuable and useful as information.
Posted on September 13, 2007 06:29 AM | Posted to Being European | Web 2.0 | Permalink
Trackback Pings
TrackBack URL for this entry:
http://www.thealarmclock.com/mt/mt-tb.cgi/5891
