TweetPsych wants to get inside your head

We spent some enjoyable time earlier this week playing with TweetPsych, a site that puts linguistic analysis algorithms to work figuring out just what's with the most compulsive Twitter users out there. Currently in beta, the for-entertainment-only analysis still provided us with some amusing insight into Twitter talk -- and into the brains of three Betanews staffers.

The site, developed by Dan Zarrella of HubSpot (home of the addictive Twitter Grader), builds a "psychological profile" of a given Twitter user based on his or her last 1,000 tweets by running the text against two algorithms that look not at what topics people are talking about but at the cognitive processes they seem to be using. The RID (Regressive Imagery dictionary) algorithm sifts texts for their primary (free-form, associative, creative), secondary (logical, problem-solving), and emotional content, while the LIWC (Linguistic Inquiry and Word Count) algorithm matches words against 82 language categories that can roughly estimate the writer's mindset. The LIWC is a widely used linguistics database; the RID is less so.

Your reporter is one of those people who stubbornly refuses to read her daily horoscope, as her day is already filed with enough vague bunkum. (I'm in the media, after all.) Still, getting to know my colleagues via an online application is much more convenient than having conversations with them or even reading their tweetstreams, so I happily shoved a couple of co-workers's usernames into the maw of the site.

Co-worker #1 -- no names, let's call him Tate -- is actually a mediocre candidate for TweetPsych analysis, as he has tweeted just 703 times. (The service works best with at least 1,000 tweets to crunch.) Still I persevered. According to TweetPsych, that's what Tate would do; the analysis describes him as someone who tweets more than the average person about work, cognitive processes, and the past and future. He's also apt to post insights and use similes in his tweets. He makes a lot of temporal references -- typical publisher -- and engages in a lot of abstract thought and constructive behaviors. He thinks about glory and moral imperatives, and the guy's a little anxious.

Amused, I looked at another staffer, whom we'll call Jim. Jim is a slacker -- not according to the analysis, but according to his pitiful history of just 41 tweets. (Sorry, Jim, I know you probably have a life or something.) I threw him in anyway, and TweetPsych did what it could; it tells us Jim also talks a lot about time and motion, along with sensations, particularly audio and taste sensations. (It also says he has an oral fixation, but that most likely comes from tweeting about food and cigarettes and sore throats.)

And your reporter? I swear a lot, but that's not news. I talk a lot about cognitive processes, speak more of the past than of the future, use a number of similes and metaphors, and tend to tweet abstract thoughts more than items of temporal import. I'm more social than my co-workers (on Twitter I certainly am, with 1,687 tweets so far) and I'm more likely to express affection or positive feelings -- when I'm not swearing, and sometimes when I am.

The analysis also delivers a short list of other Twitter users who "think like you." I found its results remarkably useful, giving me five remarkably well-chosen follow recommendations. Interestingly, one of the five recommended Twitter users is a friend of a friend with whom I lost touch years ago.

The head-shrinking was all in good fun, though nothing I'd take much more seriously than those unread daily horoscopes. But a recommendation engine so specific and fine is worth further scrutiny. Twitter has experimented over the past few months with ways to get new users following high-profile Twitter users they might enjoy, even as reports indicate that most users try the service once or twice and drop out. Right now TweetPsych works best with a frequent user such as myself, but what if new users could write a paragraph or so about themselves and get back highly personalized suggestions? Seen in that light, TweetPsych is an intriguing glimpse into one possible direction for the highly hyped, yet weirdly underutilized, microblogging service.