<< A Century in Photos | Front Page | Faith-Based Economics >>
Thursday, November 20, 2008
Thinkers We
Teh Typealyzer is one of those interweb thingies that classifies things when you feed it some this or that. In this case, you feed it a URL and it tells you what kind of blog lives there. I fed it The Valve’s URL and it deduced that we are Thinkers: “The logical and analytical type. They are especialy attuned to difficult creative and intellectual challenges and always look for something more complex to dig into. They are great at finding subtle connections between things and imagine far-reaching implications.” I especially like that “far-reaching implications” part. Further, we “enjoy working with complex things using a lot of concepts and imaginative models of reality.” OK.
Teh Typealizer also coughs up a diagram of the brain that’s shaded to indicate which parts we use when thinking Valvish thoughts. Color it worthless.
As for how it does what it does, David Beaver at Language Log guesses:
The Typalyzer [sic] seems to use a standard machine classification technique. Presumably the creators have created a little database consisting of example websites with classifications that they think are appropriate, then trained the classifier with these examples. Now the classifier gives instant classifications of sites it’s never seen before based, broadly, on how similar the new sites are to the old ones. It’s likely that the classifier is primarily using what’s called a “bag of words” technique. That is, similarity of one site to another is based on the relative frequencies of words used on the sites. Then based on the classification, the machine chooses which misleading picture to take from a vast database of misleading pictures. All well and good. If you’ve any faith at all in the psychological categories they use, then this is neat way to do cheap psychology.]
Seems reasonable to me.
Comments
I don’t think the brain picture is at all to be taken literally - in fact, I didn’t even notice it behind the axes of the graph until David Beaver’s post pointed it out. I highly doubt anybody’s under the illusion that there’s some physical mapping that corresponds to the T/F or S/N polarities.
I actually think the tool could be quite valuable, though I wish the creators would divulge some specifics about their methodology so we know how much human intervention there is in tuning the classifications, and at what stage it occurs.
Like some of what goes on in literary theory (such as the continued influence of Freud), the “correctness” of the Myers-Briggs as a psychological model that corresponds to reality doesn’t matter: what matters is that many people use it as an intuitive form of representation anyway, and a machine’s capacity to emulate that representation can tell us a lot about the way people read voices - like the kind of formal features they pick up on to infer the persona behind a text.
It may not be scientifically rigorous, but it’s good fun. As the previous poster says, the classifications seem to be from the [url="http://en.wikipedia.org/wiki/Myers-Briggs_Type_Indicator"]Myers-Briggs Type Indicator<A>. A straw poll of blogs gave about the resuts I’d expect:
INTP: The libertarian lawyers over at <a href="http://www.volokh.com">The Volokh Conspiracy[/url]
ESTP: Village Voice sex and gaming columnist Bonnie Ruberg
ISTP: Those Godless Dawinists at Pharynguala
I’d be interested to know how they trained the classifier. The obvious way to do it would be to use a set of blogs whose author’s Meyer’s Briggs type was already known (e.g. you ask a sample of bloggers to take a standard Meyers-Briggs test).
It might be even more interesting if they’d trained it against MMPI-2 scales… (Do Bonnie’s contributers score high on measures of depression like scale 2/D or DEP? Do Volokh’s score high on scale 4/PD? Hours of fun.)