Comments on: Visualizing all the broken hearts

By: watz

watz — Sat, 18 Feb 2006 09:55:31 +0000

I meant more like data flags, like the sex and age fields you show right now. There is something appealing about a clinical analysis of romantic issues, like the stuff you mention Nigam’s classifier doing.

By: golan

golan — Sat, 18 Feb 2006 00:39:43 +0000

I assume you mean interface choices for those tags.. like being able to filter all the posts by those that involved cheating for example. I’ve considered this seriously but mostly resisted the temptation. I’m concerned that it would turn the project into too much of a control-panel. With pull-down menus, checkboxes and radio buttons in an already hectic interface.

I am going to do some more work using color and size to better articulate the underlying information. I’d also condider floating diagnostic ‘help’ balloons that could be turned on and off.

By: watz

watz — Fri, 17 Feb 2006 11:07:52 +0000

Thank you for the clarification, Golan.

Flags for “cheating involved” and other characteristics would be excellent. Looking at the text excerpts, it seems to be quite common… The similarity search is a bit hard to understand as it stands.

By: Golan Levin

Golan Levin — Fri, 17 Feb 2006 07:07:13 +0000

The project is still under development and had to go up before all the features were in place. The basic idea is that the currently selected breakup (the yellow one) acts as a search into the complete set of 20000 breakups. All of the other breakups then re-color themselves according to their similarity with the selected one. Similarity is judged according to a weighted combination of a lot of different properties. Many of these properties have been tagged by the software of my collaborator Kamal Nigam, whose expertise is in natural language processing. His classifier attempted to tag breakups with inferences about the emotional state of the author, whether cheating seemed to be involved, etc.