Sunday, October 9, 2016

Hacker Factor: Automated Gender Assumptions

Can algorithms hone in on a quality like gender? How accurate are forensic experts who analyze emails, letters, ransom notes, and the like to create a writer profile (age/gender/education level/etc.?)

I first came across the Hacker Factor online website when it was mentioned in a video discussing the unusual case of Karin Catherine 

In an effort to determine whether the posts on Karin's profile were in fact written by a female, samples of her writing were copied, pasted, and submitted for automatic analysis on Hacker Factor. The results came back as having most likely been written by a male. The woman behind the FB video had the same results. I've now joined their ranks.

The site gives you a (potentially arbitrary) 300 minimal word limit on the writing samples sent in for analysis. You can of course submit shorter samples, but the site warns that the longer the sample the more accurate the result may be. In addition, the site initally cautions that there could be inconsistencies in the analysis when looking at informal vs. formal writing vs. a combination thereof, or whether a female's writing sample (say, a news article,) was edited by a man.

First, I analyzed one of my most recently written poems, "This is Not a Dirge." It determined that informally, there was no chance I was a female, the results stated 100% male. Interestingly, a formal speech analysis indicated a "weak female" result, and further it presumed that my style weakly suggested a European background. 

I put other pieces of my writing through its faceless scrutiny - including a portion of one of my short plays, an art criticism article, and a recent article I wrote about Prince (particularly about my love for him and his countless philanthropic efforts.)

One of the outlier results of these analyses occurred when I submitted text from an informally written Facebook Messenger conversation regarding my budding research into the Chris Dorner case and my interest in the fascinating content of his manifesto - how well that would translate into a script and how interested I was in further exploring this material and perhaps writing my own manifesto (an edgy and angst-filled but accurate roman a clef) parts of which could be used as monologues - kind of like post-mortem gonzo method acting journalism plus mental illness theatre. 

I saved this conversation, this passage I wrote to a particular individual, because the person in question seemed ready to pounce on my idea and essentially steal it, a claim backed up by a recent FB post wherein this same individual now suddenly claimed that they were now looking into the case and manifesto. 

That's funny to me, though not in a "ha-ha" kind of way because this person never once mentioned Dorner's name in conversation, much less as an idea for a script or story, and I've known this person for nearly a decade - certainly since before the Dorner incident even broke. If you're reading this, you know who you are. I don't play games, nor will I be polite enough to spare you the embarassment and omit your name next time. Stay in your lane. Take this under strong advisement, so to speak. 

(I've posted screenshots of several of these results above.)

Hacker Factor is an interesting concept and I'm itching to know what criteria is used in the analysis. Even when I purposefully included more heartfelt passages from my Prince article my results (especially in the "formal writing" analysis) concluded male. 

The site boasts of an over 50% accuracy rating. Not in this gal's case, but an intriguing tool nonetheless.

I'll be looking into this site some more and posting a follow-up. For now, I hope your curiosity is sufficiently satiated. :)

To check it out for yourself, visit: