LexaBlog: Our Sentiment about Text Analytics and Social Media
Maybe the end of the world is a good thing (well… for some)
Given the continuing stream of bad news that assaults us each and every day… 5000 layed off here, 8000 there, and the political parties battling over what shade of lipstick to apply to the pig, you wouldn’t be alone in feeling down and uncertain. There are however some bright spots out there, and thankfully we seem to be dead in the middle of one of them.
There was a very interesting article in PRWeek that we came across today discussing the success that a number of the vendors in Social Media Monitoring are having in this economy, and thankfully some of them are our customers. While this may come as a suprise to some, its something I expected to see, because one thing companies can’t afford to do in this economy is ignore the bad news. They need to know what’s being said about their companies so they can minimize panic and damage around their brand. Of course this means they have to look into more places (blogs, twitter, facebook, etc) and have less people to do the looking (remember the 5000 layoffs at the top of this post), so they are having to look to companies like dna13, Evolve24 and Cymfony to do the digging for them. The PRWeek article spells out how well many of these guys are doing in this tough economy.
My general opinion is that the companies that help companies handle the ever increasing glut of information that’s out there are going to thrive in this economy, and thankfully we’re one of those.
- Jeff Catlin's blog
- Login or register to post comments
Comments
Great post, we’ve also found it pretty easy to build a system that gets 70-80% accuracy in almost no time at all.
Sentiment accuracy is also interesting and measuring it becomes all the more difficult when you consider it as a spectrum rather than simple agree/disagree.
For example, when dealing with financial sentiment measurements as we do, you can be a little wrong, or you can be REALLY wrong.
Getting the right polarity 80% of the time is great, but you also need to consider what 20% you missed. Humans who disagree will usually have agreement on the highly polarized articles. We expect people to disagree on the more intricate cases.
In our experience, even if you are getting the same % agreement overall human-human or human-computer, computers are much more likely to throw articles humans would all agree on into the wrong bucket.
Thanks for the feedback Bryce
You make a good point about ranges, and it would be something that I would love to test with some external data, but I’m just not aware of any good sources.
Internally the model based system supports a range of ’sentiment classes’ rather than just Positive / Negative and of course our phrase based model produces actual scores so the higher (or lower) the score the more or less posivitive it thinks a document is.