2007-02-27

Bye, PC... Sony Vaio Notebook for Sale.

The transition period is over and my switch to Mac complete. I don't need my PC notebook anymore and I am offering it on ebay here. Speaking of ebay: I have received three mails from people asking me for an instant buy option (at ridiculously low prices). Seems to be a new strategy for bargaining low prices instead of bidding? Which would explain why 18 people watch the thing and 0 people bid. Weird.

Labels: , , , , ,

2007-02-15

Anarchic collaborative LED art

What a neat idea: Instead of vandalizing the walls next to your favorite section of your train ride to work/university/whereever with pointless graffitti, you can you channel your artistic aggession into something much more subtle: LED throwies! The Web 2.0 version of grafitti. An LED throwie is a little glowing LED with a battery and a magnet attached to it. You can build one yourself (instructions here) for less than 1 EUR apiece. The idea is build hundreds of them, to wait until nighttime, grab some friends and toss them to metal surfaces. Anarchic collaborative LED art. Batteries last for 1-2 weeks depending on the color of the LED.

Labels: , , , ,

2007-02-10

A common subgraph approach to knowledge network comparison

In my PhD - thesis, I develop a computer-based psychometric test: the Association Structure Test (AST). For every subject, the AST elicits a graph representing the subjects' knowledge structure for a specified knowledge domain. The AST is based on concepts by Schvaneveldt, Jonasstn, Yacci and Beissner, Eckert and Bonato. The aim of the AST is to predict individual performance in intellectual tasks with graph-theoretic concepts derived from the subjects' knowledge graphs. A typical knowledge AST-elicited graph looks something like this.

(This participant was a cleaner that completed the AST with the stimulus “carpet cleaning”)

The problem is that the graph-theoretic measures I employ (density, size, degree, relative maximum path length and clusters) do not correlate with outside criteria under all circumstances - sometimes they do, sometimes they don't. I am thus looking for a better measure and the idea is to develop a kind of smilarity measure for comparing subject's knowledge graphs to an expert's reference graph for performance prediction. This has been done before in Pathfinder-based approaches, but in these studies, all networks had the same number and type of nodes because nodes represent knowledge concepts that had been selected by experts prior to knowledge elicitation and the subjects only had to do pairwise comparisons between expert-selected concepts. AST-elicited networks employ subjects' free term associations as nodes and thus all graphs differ in size, degree and node label. The idea is to apply similarity algorithms that determine structural similarity between two graphs regardless of node types as employed in chemistry for proteine comparison, e.g. a common subgraph approach as suggested by Le, Ho and Phan. The idea came up in a discussion with Gunnar yesterday (thanks to Hans for introducing us) and Gunnar was so kind to offer to do the calculations for which I am very very greateful.

Labels: , , , ,

2007-02-07

time to go

Ok, it has been on engadget today, but this gadget is so i-want-one that I can't resist to write something about it. And it has been conceived here in Berlin (at the UdK)!
The idea is so simple: a watch that knows where you are, where your next appointment will be, how to get there and how much time it takes to get there. The JITWatch gets the data (GPS position, local transport time tables and appointments) via bluetooth from your mobile (given that you have a working GPS dongle, a GPRS/UMTS flatrate, an always-on-attitude and coverage) and tells you when to leave and where to be. Still very prototype and maybe a bit too Stuttgart in terms of engineering, but still.. i want one. By Martin Frey.

Labels: , , , ,

2007-02-06

visualizing search context

exalead.de allows users to add context to searches (only in german so far). A search result also shows you associated terms, and you can select some of them to include in your search and exclude others in order to refine it. From a user interface perspective, it is an HTML version of the skillMap's graph-based exploration of a tag's surrounding. However, in the skillMap, the context of a tag is set by the community, whereas in exalead, the context is relies on automated text mining. This approach is also called "moderated search". Kartoo.com would thus be somewhere in between the skillMap and exalead, because it also displays context based on automated text mining, but does it in a graphical way. One cool thing about Kartoo is its text extraction algorithm which tries to give a brief summary of the page you MouseOver on the left. For the skillMap homepage, this works a little better than the topicalizer (read my previous post on the topicalizer here).

Again, thanks to Andreas.

Labels: , , , ,

automatic tags vs user-generated tags, pt2

Can tags be clustered automatically and brought into some form of hierarchy?
In our skillMap approach, we rely on the crowd intelligence of a/the community to network tags manually into a weak hierarchy. However, there is a discussion going on whether it is feasible to do that automatically, for example here:
www.connotea.org/search?q=tag+clustering
www.rawsugar.com/www/2006/20.pdf
www.bibsonomy.org/search/tag+clustering

(thanks to Andreas)

Labels: , , , , ,

2007-02-05

2zimmer

Ach schön. o2 schmiss mal wieder ein umsonst-konzert ("o2 music flash") und daniel und ich waren schnell geneug mit der antwort-sms, um noch ein ticket zu ergattern. Also gings am freitag den 02.02. ins spindler & klatt. Strömender regen, lange schlange, brechend voll. Ich finds bei diesen o2-konzerten immer etwas blöd, dass es keine vorgruppe gibt - der beginn ist dann immer etwas unvermittelt. aber es war wie immer großartig - 1:45h gespielt, tolle sachen auf dem neuen album und eine live-version von freie liebe die so elektrogeschrammelt daherkam, dass sie echte clubqualitäten besaß. seht selbst.

Labels: ,

2007-02-02

automatic keyword generation/tagging

A lot of the hype in the web2.0 world is around the concpt of tagging. Tags are attached to objects by human beings, but there are thoughts whether this can be done automatically. One of the first services that offer free public auto tagging is the topicalizer. It auto-generates keywords for any URL along with a load of other statistics (e. g. readibility etc). It didn't work very wellfor the German sites I entered, but it as a nice API. I would really like to do a study determining the similarity of automatically generated tags and user-generated tags (i.e. with delicious tag clouds or skillMap-tags).

Labels: , , ,

don't beleive the hype

The Hype machine dives into the long tail of obscure blog entries on new music, take a slice of the tail and processes it to identify upcoming music hypes. Hype machine is the winner of the most recent mashup camp at MIT. (thanks to Matt Stevens)

The Hype machine about itself:
The Hype Machine keeps track of songs and discussion posted on the best music blogs. Easily listen, discover and buy songs that everyone is talking about!
GigeOm:
The Hype Machine combines blog posts from a set of curated music blogs with Amazon sales data and upcoming events. It includes an amazing hacked integration with iTunes that takes you right from the web page to the track you’re interested in. If you prefer buying through Amazon, The Hype Machine figures out what CD page to display.

Labels: , , , ,