Approval vote graph

homeblogtwitterthingiverse



Consider a graph of this site based on the approval vote in the left sidebar. There would be two kinds of nodes: pages (such as blog, aether, cats_and_dogs) and words (such as elegant, confusing, broken). Pages would be linked to words based on the results of the approval vote (link strength being proportional to number of votes).

Run a graph layout optimizer over this graph.

Words often used together to describe pages would be placed close together in the layout, as would pages that are described by similar words. The really cute thing is that, even if two pages have completely different words describing them, if the words are semantically related the pages will be placed close together.


How about we take this to a grander scale? Set up a system that lets people vote on any web page on the web. There would be two votes: what it is about (eg "open source", "science fiction", "lions and tigers"), and what it is (eg "technical", "interesting", "crazy", "spam", "pictures", "trustworthy").

A graph based on the "what it is about" vote would let you browse by topic. You could use the qualities to look for pages of a particular type within a topic (eg technical or introductory/overview or well written).

You also get from this system a set of semantic relations. For example, you could infer a strong link between "open source" and "free software", or between "nifty" and "elegant", based on the fact that there are many pages having both properties.

The graph layout thing lets you extract a parameterization of the space. For example, from a two dimensional graph layout of page qualities you might find a dimension for clarity of writing and a dimension for interestingness of content.




[æ]