TechTip: Watson APIs - Personality Insights

Analytics & Cognitive
  • Smaller Small Medium Big Bigger
  • Default Helvetica Segoe Georgia Times

So far, we have been talking about Watson from a business point of view. But personality insights? Sounds mighty personal. Are you ready to be analyzed by a machine?

I don't care who you are, what you have done in your life, or how you view yourself or others view you. No one is more afraid of a personality insights app than I am. I have worked very hard to project a "normal" persona. And now I am to have it undone in an instant just to prove Watson is smart? I think not.

But no worries. It turns out that the Personality Insights API is directed more at the people who are connecting with you (your customers, your vendors) than it is at you. So for the moment, I am OK.

The idea here is to give you an idea of your contacts' needs, values, habits, and preferences, based on what they have written to you.


By now I suppose you know the drill. Go to the Watson home page. Click on Products and Services. Then click on Personality Insights under the Empathy heading.


In short, this API uses linguistic analysis to identify the "Big Five" personality factors. The input for this analysis will be emails, blogs, notes written on a cocktail napkin, and forum posts. The output is designed to show preferences in terms of entertainment, shopping, and music. Yeah, this may be more for entertainment and retail customers than your traditional ERP organizations. Just a guess, but that's what I think.

The Big Five

Five main personality areas are rated:

  • Agreeableness - This measures a person's tendency to be compassionate and cooperative with others.
  • Conscientiousness - This is about a person's tendency to act in an organized or thoughtful way, their tendencies toward extraversion (seeking stimulation in the company of others), and emotional range (the extent to which a person's emotions are sensitive to the individual's environment).
  • Openness - This is the extent to which a person is open to experiencing new activities.
  • Needs - This includes excitement, curiosity, harmony, closeness, self-expression, liberty, love, practicality, stability, challenge, and structure.
  • Values - The values we are talking about here are not things like stealing or believing in monogamy. Instead, we are talking about more subtle characteristics, the kinds of things that motivate us to do or buy certain things.
    • Self-transcendence (helping others)
    • Conservation/tradition (self-restriction, order, resistance to change)
    • Hedonism (seeking pleasure for the self)
    • Self-enhancement (success for the individual)
    • Openness to change/excitement (independent action, thought, and feeling while being ready for new experiences)

In other words, you have to be both structured and open to anything. Yeah. Good luck with that. The overall score provided may (or may not) help lead you in that direction. My experience is that people are either open or not; they can't be both. But IBM may have different experiences.

How It Works

In one sense, this product seems like a consumer's nightmare.

The basis for Personality Insights is social messaging: emails, texts, forum posts, and tweets. It takes that information and makes projections of your likes and dislikes that can be useful to businesses to help them tailor their message to you.

It appears that the analysis returned is more general than specific, showing certain trait tendencies rather than specific likes and dislikes. Unless you tweet excessively about how much you love Coke, that fact is not likely to show up on your profile.

There is a demo feature that includes five or six people and a set of data attached to each of them. By selecting a person and running Analyze, you get a short profile. It's kind of fun to play with, and it's nice in a way to know that the Pope is "likely to volunteer for social causes" but "unlikely to have a gym membership." Oh, and he likes entertainment magazines, too.

What You Need to Do

Working this into your system seems relatively easy. Personality Insights is activated when you call the Profile method. You can pass it up to 20MB of social media-type data, but the latest release of the method allows you to get away with much less data and still get an accurate profile.

What You Get Back

The output of the Profile method comes in two formats: CSV or JSON. Which you get depends on the parm you pass in with the ACCEPT header (text/CSV or application/JSON). There are of course format differences, depending on what you use.

In general, this is what you will get back: Personality, Needs, and Values information is returned as an array of Trait objects. Behavior is returned as an array of Behavior objects, and Consumption as an array of Consumption objects. The behavior objects are returned in a date and time format but only if the input data was time-stamped.

For Personality, we are looking at Openness, Extraversion, Conscientiousness, Agreeableness, and Emotional Range. For each of these, you will get two scores between 0 and 1 (raw and percentile). Raw speaks for itself; the higher the score, the more likely you are to have that characteristic. The percentile is based on how you rank against a standard population. So if your percentile score is .6549388585767584, then you are in the 65th percentile and are better at something or more into it than 64 percent of the standard population. What I am not sure of is what this "standard population" is. Is it all of the data that you have entered for various individuals, or is it some national or international standard? Doubtless, IBM can answer that question if you are interested in this app. For more information on rankings and to see the actual objects, check out this section on the site.


Doubtless, this app will be evolving as there are many areas other than retail where this information could be put to use. I'm not sure I like that, but then I'm still hoping to spend the last few years of my life living in isolation in the high Sierras, so I may not be the best judge.