I've complained that the web doesn't have a memory and you can't do searches that are limited to certain time ranges, to analyse the rise and fall of certain topics, say.
I had thought that this would be particularly nice in Feedster, to enable us to see how particular topics spread across the world or the limited, incestuous blogdom at least. That way it could be easy to spot, track down and shoot in the back hypes – or memes as they're called these days.
The Recall search engine seems to allow such searches now, even trying to give some sort of context. Not exactly what I had in mind, but definitely the technology to implement my ideas exists.
Another view on the history of documents gives the history flow project by IBM. Nice.
Ok I’m game. But can you give me some thoughts on user interface? Email them to my feedster address — scott A[T] feedster.com and I’ll try and figure out how to get it in there.
. I guess the most obvious and straightforward thing would be to simply draw a graph of occurences per day. That could be quite interesting to begin with. For example you could see what took people by surprise (I suppose searching for iSight would give a steep peak that decreases exponentially) whereas searching for anticipiated events (Halloween, say) will give something like a Gaussian around the event.
. In a next step the graph could be an image map that lets you click it and will restrict search to the period of time associated to the click. When searching for ‘iMac’ say, you’ll probably get peaks for the different revisions of the machine. Using ordinary search engines, you’ll mostly end up with results on the most recent revision only - which may not be what you’ve been looking for.
. Metadata permitting, this could probably be taken even further by incorporating geographical information (which percentage of blogs uses GeoURL or similar tags? Any idea?): Have a map and draw a blob for the occurences of the search term. Use a colour gradient to indicate time. That way it may be able to see how (or if) some news spread across the world.
. The service I mentioned seems to go even further, by trying to find context for the search term (although I find their UI a bit non-descriptive and confusing). By looking at the terms that frequently turn up in the same posts as the search term, you can probably correlate them and find out the context. I guess making this work well will probably require deeper changes to the database to have ready access to the correlation information.
Of these points, I think the first two are probably both the most useful and easy-to-implement. Lucky coincidence.
Received data seems to be invalid. The wanted file does probably not exist or the guys at last.fm changed something.