Thursday, October 18, 2012

I'm not negative, you jerk

Hey, remember how I used to do these random word clouds of my blog? No? Cos I haven't done them in about a year? Right, makes sense.

The other day I remembered I hadn't done one of those in awhile and decided I wanted to again. But whenever I do build one it's just made up of my 6 most recent posts. But I want my word cloud to represent MOAR DATA. So I decided to play around with a social listening tool to pull in more posts. Here's what that looks like

Yeah I know, not that interesting. Oh I say "book" and "read" a lot? Who would have guessed.

BUT while I was in this tool I decided to take a look at the sentiment section and see what it says. And it told me I am mostly negative. Wait, what?

It said, for the past year, 22% of my content as been positive, 37% neutral and 41% negative. Granted, my first reaction to those stats was to IM a friend and go "Apparently my blog is 41% negative. Well fuck you then!". So yeah, maybe they have a point.

Then I really looked at why the system was being so judgy. Essentially anytime I used a negative word like "bad" or "don't like" it said the section was negative. Of course, as with any automated system, it can't take into account context. The system can't tell I said "Now normally I don't like this, but this book nailed it perfectly. Kudos" It also marks my entire rant about wanting to set a character on fire because I hate them so much as "positive" because I said the character was "SO PERFECT".

Moral of this post:
Word clouds are pretty boring but have shapes and colors so people like them (me included)
Context is important
Computers do not get sarcasm
Automated sentiment is pointless