I installed a widget that allows php code to run and stuck the Top10 script in there. The data is reasonably accurate but sometimes my collect script stops for a few hours at a time. In other news I’m running into some kind of error in Tag Helper, an array out of bounds exception. It may be the case that my sample size is too big because other documents run through the trial fine.
This post has nothing to do with Thanksgiving but I’m running out of post titles. I’ve been doing most of my updates through Twitter because it is very convenient but I realize I need a real update. Since last week I have collected 30k+ users and 40k+ tweets. Early last week I got an interface for WEKA called Tag Helper and started playing with that. My initial thought on rating posts was that there would be some type of scale, maybe 5 ratings. It has come to my attention that most of the time these sorts of things are done with just two ratings, and if more are needed then you do layered filtering. So if you wanted Positive, Neutral, Negative it would be a two stage process of separating out Negative from Not Negative and then filtering out the Positive and Neutral from the Not Negative set. So at this point with the project due dating drawing closer I’ve decided that to keep in line with the main goal of this project I need to settle for something that works rather than what a full development team might end up with. So I think I’ve chosen to pick out posts that are distinctly negative from the rest. This set is particularly interesting to businesses because, at least the case with Starbucks, most negative comments tend to be packed with suggestions or alternatives to current practices. If I could not only identify what % of tweets are negative, but pick out trends within those posts it would serve as a digital suggestion box. I have 750 tweets labeled, but have not run it through anything yet. That will be my goal for this week along with starting the paper.
