Gender Analyzer: did a man or woman write that blog?

The folks who developed Gender Analyzer say their web service uses artificial intelligence to determine if a website was written by a man or woman. According to the 'bot, the front page of boingboing.net was written by a dude, which is about 3/4 correct! Jokes aside, it looks like their rate of accuracy in general is pretty good. (thanks, @sandroalberti!)

• [The Tensor]

  Another data point: it thinks my blog was written by a woman (and I’m a man). Accuracy currently holding steady at 56%.

• [John Mark Ockerbloom]

  I fed in the URL for A Celebration of Women Writers, which is not only edited by a woman (my wife) but is all about women’s writing.

  As of the time I write this, Genderanalyzer says “We think [this page] is written by a man.”

  Go figure.

• [Sekino]

  I’m just unimpressed that their ‘artificial intelligence’ is equally dependant on generalizations and prejudiced assumptions as the average human to make its call. Actually, I’d bet most people would probably have a greater rate of success guessing the blog writer’s gender themselves.

  Do we call it ‘artificial ignorance’ then?

• [Zan]

  It’s accuracy is less than 60% now, which isn’t much better than a coin toss.

• [Cool Products]

  It surprises me a little that there is only a 68 percent probability that I am in fact a male.

• [arkizzle]

  According to the other version of this over at http://bookblog.net/gender/genie.php (linked by ZIIB @ 11), some of the keywords involved are thus: (I can only presume they use a similar algorithm)

  FEMALE:
  [with]
  [if]
  [not]
  [where]
  [be]
  [when]
  [your]
  [her]
  [we]
  [should]
  [she]
  [and]
  [me]
  [myself]
  [hers]
  [was]
  [to]

  MALE:
  [around]
  [what]
  [more]
  [are]
  [as]
  [who]
  [below]
  [is]
  [these]
  [the]
  [a] 3
  [at]
  [it]
  [many]
  [said]
  [above]

  (Teresa, ‘scuse the verticals, BB ain’t fixed-width)

• [Trent Hawkins]

  to Zan: Yes, it’s basically a 50/50 coin toss. However it is an advanced virtual reality coin toss that warrants thousands of dollars of research funding.

• [jccalhoun]

  I tried a few gaming websites written by women and it said 3/4 of them were written by men.

• [Secret_Life_of_Plants]

  The computer classified my two blogs, and the blog of my roommate as having been created by men. (I am a woman and so is my roommate).

  I will say that even on message boards, people responding to something I wrote inevitably address me as “dude” or “that guy”.

  Oh well.

• [LYNDON]

  Interesting. I just put two of my blog posts and a satire column through the gender genie, and I turned into a female for the last one.

  …which was a parody election manifesto, so to was the “we”s that did it.

• [retchdog]

  Yet another boastful claim based on performance on the training data, blown to bits when exposed to the world’s diversity.

  The map is not the territory; especially if the map is drawn by some dot-com-2.0 yahoos with too much python and not enough statistics.

  Unexpected bonus: With seven tabs open, my firefox truncates this one as “GenderAnal…”

• [Jerril]

  Currently, the correct rate is 56%, which is uncomfortably close to “we flipped a coin”. They don’t give any information on sample size on the poll/result feedback, and of course there are no controls on responses so who knows what people are saying…

  But yeah. Currently, totally pointless.

• [Pam Rosengren]

  I guess I shouldn’t feel bad about it deciding I am a guy.

  But having grown up in the days when we were taught in science class that women’s brains stopped growing before men’s, that an intelligent woman was an anomaly and that intelligent women tended to have smaller breasts and be less attractive, this analyzer makes me boak.

  If they are researching whether there are any statistically significant differences in the way each gender writes, and what we write about, their results at this stage are “not really”. If the researchers have assumed there is a difference, then the results so far suggest that they need to revisit their assumptions.

• [Anonymous]

  I’m willing to bet it’s just a standard Bayesian filtering mechanism. So, for example, tech writing on the web is dominated by men (and easily identifiable) and it’s just trained to classify accordingly. Don’t read too much into it!

  - Udo

• [macguffin]

  Well put, Pam. It’s neat when it gets it right, but that seems to happen only half of the time. The generalisations just don’t work.

• [gabrielm]

  “Could not find female probability in response…”
  I hate when that happens.

• [erissian]

  To be fair, anything above 50.02% is an improvement over just guessing; even if it isn’t by much it’s a start.

  @#18: This /is/ training data for it. It will become accustom to the world’s diversity, and on June 9th, 2012 it will become self-aware.

• [IamInnocent]

  Apparently it is a little confused now… maybe someone asked it the question about itself and their server is kept in a closet ?

• [Sekino]

  Every web paged I’ve checked gets a “male” output. I’d like to know the algorithms behind this.

  Keywords relationship, shoes; animated GIFs of sparkling fairies, kissing emoticons = female

  Keywords gaming, beer, technical lingo; animated GIFs of large breasts, mooning emoticons = male

  That’s my guess. Clearly, we’re dealing with geniuses. :P

• [Lauren O]

  Clearly, we’re dealing with geniuses.

  Hey, these people know their gender stereotypes very well.

• [justanotherusername]

  Who cares anyway?

• [rolandog]

  Drats. It says I’m neither a man or a woman… I seem to have a gender error.

• [tubgull]

  @1

  Could be worse. Last time I directly inquired as to someone’s gender, I did not get so gracious a response.

  Maybe someone could make a similar website that lets me ask women if they’re pregnant.

• [Muscato]

  Three errors – and then being taken for a woman. Takes me back to the 80s, when at least I had feathered hair and too much eye makeup…

• [LYNDON]

  My guess the kind of thing it’s looking for would be standard language features – as opposed to content – that are statistically more used by one gender.

  In spoken conversation these usually include things like, kind of, more hedges and, um, conversational noise, y’know?

  (To my mind the gender trait here is signalling social non-dominance, which the females do more)

  Or using “I think” rather than just asserting your opinion as fact. Or empathising and being agreeable.

  I know someone did a thesis looking for these kinds of thing in IRC-style chat – specifically at whether people moderated their language style to the supposed gender of the person they were talking to.

  This is statistically real diffrence but you’ll find a lot of overlap between boys and girls.

  But I don’t know about more deliberate written work – people will use different voices for different purposes, For eg, as noted above, technical documentation will come out male.

  Of course it’s (I imagine) training and been fed a lot of examples – like a clever spam filter. So if it gets to working, it would indeed be interesting to see how.

• [Nancy Jane Moore]

  Well, it thinks my self defense blog — Taking Care of Ourselves is written by a man. I can’t help but think that the concepts and language of self defense are automatically typed as male in our society.

  Or maybe the program just latched onto the words “fourth-degree black belt” and assumed I must be a guy. (In case you didn’t look at my name or assumed “Nancy Jane” was just my online presence, I’m female.)

• [Abby]

  Cuteoverload isn’t written in English, apparently.
  “Sorry, we can only classify web pages written in english”

• [Sister Y]

  My theory is that it thinks you’re a woman if you talk about your feelings and stuff, and a man if you don’t.

• [Anonymous]

  All technical articles written by me (female) are classified “written by a male”.
  Only my English homepage where I state clearly “gender: female” is classified correctly.
  Either not working for female geeks or its confused by my non-native English. Or should a female care about other themes? Then this tool is a case for real gender studies …

• [Clumpy]

  Every web paged I’ve checked gets a “male” output. I’d like to know the algorithms behind this.

• [Mina]

  I put in all the blogs I write for the literature and film courses I teach. Everyone of them came back “Man.”
  I’m not.

  I’m guessing they were labeled that way because of the kind of language they’re written in and possibly the topics discussed: intellectual discussion and analysis couched in theoretical, technical and philosophic discourse.

  In other words, crap gender stereotypes.

• [zuzu]

  There are no girls on the Internet.

• [Nytespryte]

  It had me as male and my boyfriend as female.

  No one was especially surprised by this.

• [Antinous]

  Apparently gender differences are pretty damn important to them. There was a “Yes on 8″ ad on that site.

• [Ziib]

  Here’s a much better (or at least transparent) gender algorithm: http://bookblog.net/gender/genie.php

  I prefer it, as you can select individual blocks of text, rather than pointing a robotic Internetbeast at a single web page. Plus it shows you *why* it thinks you are irrevocably female, or male, or lord-knows.