MixedContent

I recently found out about echonest. Don’t bother clicking. There’s nothing really there yet. But both founders Brian Whitman and Tristan Jehan seem to be up to neat things. A good article in the globe and mail gives some background on their new start up.

In a nutshell, the two have been developing software that listens to music and tries to predict which songs will be ‘hits’. Unlike some other services that concentrate only on the physical properties of music (rhythm, pitch, timbre, etc.) their software also scours chat room, discussion forums, blogs, and anywhere else music is discussed in order to glean pertinent information. This is a great idea, because (as I’ve said before) music depends on the social context in which it is created.

In the past, I was thinking about using a similar approach for stock market prediction. I thought it would be a good idea to combine a strong statistical analysis of stock data (using HMM’s and NLP techniques) with some real-time data collected by bots searching internet discussion forums, news releases, SEC filings, etc. I think this would work because stock prices are in a nice feedback loop with all this online data (Both influence each other). To ignore all this great socially relevant information about stock prices, and to focus just on the ‘numbers’, is akin to operating with huge blinders on. But I digress…

In any event, I look forward to seeing what echonest comes up with!