Software Tracks Blogosphere Mood Swings

holy_calamity writes "Dutch researchers have figured out a way to measure the mood swings of the blogosphere. It can pick up peaks of flirtiness from bloggers around Valentine's Day and drunkenness at weekends, the plan is to create a search engine that returns the prevailing mood in the blogosphere about a topic. Companies are already interested in using it to track consumer confidence. What's the mood of Slashdot on this one?"

Re:skeptical

[TechnoGuyRob]

Erm...I don't know if you read the article, but they extract the moods from LifeJournal posts, not analyze the text or anything like that.

Assuming that people are honest about their moods (and why wouldn't they be?), I don't see why this wouldn't be accurate.

Apparently, your mood right now is ignorant.

Website still up?

[Chris_Jefferson]

This can't be right.. the website is still up. Perhaps that is because no-one can find the link To the actual moodviews website [science.uva.nl].

I can't decide if I should feel guilty for posting this..

Re:Been there, done that.

[daviddennis]

Actually, the article was pretty clear on how it worked, but I'll explain it a little better since I guess (from your post) it was confusing.

When you write a blog entry in LiveJournal, you're give an opportunity to select a "mood" from a dropdown list of moods. So you can say you're

happy
sad
loved
excited
lonely
(etc)

by just picking the appropriate word.

Now, as you know, emotional data taken from a dropdown list at the end of writing a blog post might not be worth taking all that seriously, but it's data, and you can try to analyze it.

Now, some people are laughing at this by saying that it should of course be obvious that people are likely to be feeling loved or lonely around Valentines' day. But actually this is an important observation, since it says that what people pick in the dropdown can be related to real events. Of course we know people are loved/lonely on Valentines' day; what we didn't know is if what they picked on the dropdown was meaningful. Now we know it is, and so (in theory) we can use this to predict events or people's behaviour based on what they say.

The Harry Potter example showed that this could in fact be done, and this means that further reasearch might be promising. For example, let's say there was a "suicidal" mood in the list. It would be interesting to track whether actual suciides were predictable or at least more likely after showing those moods, so that an early warning system for such behavior could be created.

On paper, it seems possible that lives could be saved that way, which makes this a non-trivial application indeed. To support my theory, note the previous news reports we've seen here that note that suicidal behavior was often predictable in hindsight from what people wrote on their Myspace profiles. If this could be determined from moods, which are trivial to check automatically, it might be a very interesting result indeed.

Hope that helped people's understanding.

D

Re:Mixed Signals

[patio11]

I could see this working. Is it any harder to tell if a message is, say, depressed than it is to determine if a message is a commercial pitch?* Because pure-"Bayesian" analysis of spam routinely gets 95%+ accuracy, which if we're not talking about the content of any specific message but are trying to measure trends between time periods is plenty good enough. Lets take a particular application: Apple wants to know if their iPods are still the hottest thing on the planet. Simple process: have a team of humans hand-classify 1000 posts on LiveJournal mentioning iPods (just grep for it) as "pro-iPod" or "anything else". Then have your trained classifier slurp up every post that day, discard any that doesn't mention iPod, and classify ahoy. Ten minutes later: "Beep, 87% of 23842385902 messages mentioning iPod (5.3% of all posts in last 24 hours) are mostly upbeat."

* Whether it is or not is not obvious to me, but setting up an experiment to figure out which it is isn't that hard.

Re:Blogosphere Mood

[run4ever79]

It is already on my list of overused bs words

• Blogosphere
• Web 2.0
• Netizen
• AJAX
• Podcast
• ROI
• TCO
• Mash-up