See here for a full PDF lecture on our recent work in this area.
In a recent blog, we discussed how Network Analysis is transforming the ways in which modern HR Departments go about their business. We’d like to touch on that again, but this time from the perspective of machine learning. Not that you are a machine. Simply put, machine learning is a subfield of computer science wherein programs are given the ability to learn without being given overt instruction (Google’s Neural Network, Watson, etc). At times, the realities they observe and display for us are pretty damn odd, or definitely not tasty. Given that these routines often leverage the internet to learn, you’d not be surprised that many machine learning algorithms understand reality as being composed of cats, and porn. Yeeesh. Other, more pointed efforts at machine learning have, however, yielded far more success.
One such area which, as we will discuss, is highly pertinent to organizations is the emergence of classification algorithms. Stored in various pockets around the web (and my laptop!), these algorithms are designed to identify gender, age, personality, mood, creativity, and hundreds of other human level traits from digital text (blogs, posts, Twitter, etc). A good number of these algorithmic coding processes, but not all, have been verified to correlate with more traditional and established inventories.
Why is this important to organizations? I don’t fucking know. Why are you asking me this?
No, that is a good question. We got ‘ya.
Traditionally, HR assesses employee level demographics and human capital using a wide variety of time intensive, invasive, and easily gamed surveys (the Big 5, for example). Aside from the aforementioned time intensive nature of distributing and analyzing these inventories, the real problem is that they are quite easy to cater too--don’t want to appear overly neurotic? Then don’t rate yourself a “5/5” when responding to “Do you twitch, swear, and freak out often?”
What machine learning can do is take existing data within a firm--written texts in the form of emails, internal notes, public posts to social media, or internal social media platforms--and quickly and easily code it for all the variables one may want. Awesome! And what is even cooler is that the logic driving these algorithmic coding processes rely on non-intuitive patterns and schemas. Why does non-intuitive matter? Because it’s hard to game. Of course no one is going to include tons of overt reference to being neurotic in their email (but maybe some do?), but subtle variations in neurotic v non-neurotic use of written word are quite obvious to said algorithms. What is even cooler is that with something as scant as ONLY the subject line text from emails, we can reliably and efficiently assign a wide variety of personality and demographic information to your employee population.
It gets cooler, though. Seriously.
By combining these algorithmic processes with network data you already possess in the form of internal communication, we can map demographic and personality level variables onto communication network structures. We can examine the communication networks between, say, your Engineering and R&D Departments looking for tell-tale signs of highly creative individuals, or pockets of individuals. Did you just roll out a new (and controversial) policy? No one is going to openly tell you “this sucks.” But they will exhibit signs of negativity, or frustration, in their email communications with other employees. We can find discontent, and help mitigate the effects of it, before it boils over.
If you’re interested in learning more about how these powerful analytic techniques can help realize your organization’s goals, be sure and drop us a line.