I had a good day at work today. I don’t know why. Maybe it’s because I got something done for the first time in a while.

The challenge I am currently beset by is quite a teaser. It involves getting events from a list and displaying them in handy chronological boxes, for year, month etc. It’s tricky because I might not have any events for January but loads in June. The data has gaps and clusters, a state which I’m sure has an excellent Com Sci name. Additionally I may have even known this name once.

I’ve just been reading about an interesting system developed at MIT Media Lab. Blogdex indexes blogs by popularity and by some mechanism is able to spot memes. The article I read glossed over a lot of the technicalities. It occurs to me that if such a system indexed the blogs daily, it could search for patterns in the words coming in. If the occurance counts were kept on each word, new memes could be detected this way. Evolutionary algorithms could be applied to clean the data based on human input. E.g. for each meme a list of ten or so sites could be displayed, humans could then rate these (probably just by clicking through to them) and the unpopular ones thrown out. This depends on good title writing. Enhancements to this system would include the whole gamut of pattern matching techniques from the AI community, especially anything good with phrases or, potentially, natural language. How far you go depends on how much money you have to spend on CPU cycles (and development time).

Another random thought was that I could in some way employ evolutionary alogrithms in Hypothetical. In searching rather than anything more sophisticated.

I miss being a student. I wish I could be an ideas hamster at a major software think tank.