Sunday, March 02, 2008

Finding nice bloggers - finding nasty bloggers

In recent weeks, development in using Bayesian inference on Latent Semantic Analysis of blogs has been keeping me up at nights.

From a simple engine developed by my friend Girish, I have been thinking about how one can use this information as a predictive tool and it has some promise.

Let me go back a bit.

Here is a simple application to find out the sentiment of bloggers between two competing entities. You can try it out. Try comparing 'clinton' with 'obama' and see who is getting the most positive and most negative blog posts.

You will also see that the programme also measures the extent of objectivity in the content analysed.

Because of the way we have set this application up if the comment about the candidates is neither positive or negative it produces a nul return (because of the nature of the experiment).

This experiment has moved on quite a lot and soon I hope to have a more advanced example you can try out here.

The experiment we are working on is to use the concepts identified in the blogs to monitor what subjects are growing and which are retreating and the sentiment attached to them.

The next phase is to predict where such concepts are going within a level of confidence that is acceptable.

Now comes the theoretical leap of faith. When a new concept emerges, it will be disruptive and may well provide an insight what future news will be.

Now that really will add a new dimension to PR practice.