Sentiment analysis is tricky anyway, even with thousands of words to mine for “positive” or “negative” indicators and top-notch machine learning and natural language researchers on the case. Of course, distilling a sample down into 140 characters or less suddenly makes that tricky problem much trickier – which is why sentiment analysis for Twitter is really kind of hit-or-miss.

Considering how divided opinion seems to be on the iPhone in the past few days since the launch, I thought this would be an interesting topic to use to try out two Twitter sentiment analysis systems I recently came across. The first is Twitter Sentiment, which rather than a commercial product is a research project (for a machine learning class) out of Stanford. If you look on their about page, you’ll see a link to a paper written about the algorithms they used to put this together. The tool provides overall sentiment as well as a graph over time – like this one, that shows when there have been spikes in tweets about the iPhone in the past six months:

Putting in a date range of just 6/25/2010 to 6/28/2010, it tells me that sentiment around the keyword “iPhone” is 64% negative. Here are some examples of the classified tweets:

NEGATIVE: iPhone 4 network is horrible – keeps dropping off. Thank goodness that I am still on my 14d cooling off period. #iphone4 #o2 #apple

NEGATIVE: Broke the screen on another IPHONE sheesh I need to upgrade anyway … are the stores all sold out

POSITIVE: Gave into temptation of buying the iPhone 4 Bumper. So far, so good. But $30 …. seriously Apple?

POSITIVE: New shoes, new dress, new top!!! New new new!!!!! And yes they all go with my new iPhone 4!!! Lol

Though my overall impression is that the sentiment classifications are pretty darn good, considering, there are some things that you can’t really blame it for missing. For example, this tweet is classified as negative: “i wish i got an iphone now so that i can play game. is very boring bored standing and the websites you can only go is fb and twitter.”

Another sentiment analyzer I tested out is TweetFeel, a commercial venture, and so unlike the previous one where you can read all about the nifty science behind it, their FAQ just assures you that they use “insanely complex algorithms” to deliver their results. The verdict? 54% positive for “iphone” after letting it run for about ten minutes.

Considering that Twitter Sentiment gives you a much, much larger corpus of data much, much faster, I’d be inclined to say it’s the better tool – though considerably less cute. Still, here’s some examples from TweetFeel:

NEGATIVE: Isaac & I are switching to T-Mobile android phones sooner then later. My iphone blows.

NEGATIVE: damn stupid iphone and stupid fat fingers! Yes I LOVE my job

POSITIVE: I’m tweeting while still listening to Pandora! Awesome iphone 4

POSITIVE: I love you iphone 4!

Again, kind of tough for it to catch nuances. For example, two tweets both classified as negative: “New iphone blows me away” and “Ugh new iPhone blows.”

So what’s my verdict for how Twitter is feeling about the new iPhone? Taking both of these together and reading through the tweets, I’d say… very mixed, verging on negative. Which is probably what you already knew just from the media coverage the past few days. But hey, here are some new toys to play with for those of you interested in analytics!