Thursday, November 03, 2005

Data Representation

I'm big on data representation, and I've posted on this matter before.

Being able to analyze large amounts of diverse data is growing ever more critical, especially since our level of specialization means we're going to have ever more complex foreign data to analyze.

The real key to this is adaptive displays, which is pretty obvious. From the same data set, you want to be able to extract the relevant data and display it in the most useful way for any given member of the audience.

The most obvious use of this is data compression. Extremely lossy data compression backed up by nonlossy secondary queries. In the same way that you only read the threads on a forum that you find interesting, you don't need a lot of data to tell whether something is worthwhile to you.

For example, allows you to bookmark things you think are cool. Such a system would be absolutely perfect for this niche if it let you do one thing differently:

Rate other users.

This is a private rating - nobody else gets to see it. However, what this ends up doing is passing you "important" links. For example, if I like Sue at 5, Bob at 3, and Larry at 2, then everything Sue bookmarks will be popped over to me, but in order for a Bob or Larry bookmark to reach me, it has to be bookmarked by someone else as well.

This propagates. People who Sue rates highly get "shadow points", and unless that feature is turned off, their bookmarks will occasionally reach me as well. So, if Sue likes David and Bufbug, and one of them bookmarks the same thing as Larry, it gets passed to me as a likely point of interest.

There are better ways to do it, but this way would require a fairly small change to, rather than an entire rewrite.


No comments: