IBM researcher can guess your personality from 200 of your tweets | Hacker News Reader

IBM researcher can guess your personality from 200 of your tweets

venturebeat.com

54 pointsco_pl_te29 comments

English is such a wonderfully imprecise language. The word personality seems to be used here as a synonym for "word-choice features we can extract from a text corpus".

If anyone thinks this gives useful predictors outside of the context of "what will they Tweet next?", I have a lovely bridge to sell you.

The assumption is that word-choice features correlate to personality traits. It's not such a crazy assumption. Of course nobody knows how strong the correlation is. Humans are messy and complex but also very predictable in many ways.

>Of course nobody knows how strong the correlation is.

Or if it exists at all, because it was an assumption.

Indeed.

I'd hate to think my tweets like "Black Eyed Beauty #AddAWordRuinAMovie" would be used without context to build up a personality profile.

In general, this research doesn't seem much more sophisticated than building a personality profile off of a vocabulary pulled from a very small subset of language usage.

I was wondering similar things: does it still work if 80% of my tweetstream is me posting #overheardAtWork / #qooc quips? Because I'm going to guess no :P

how wrong could you be. Just because you can't see the correlations, or have no intuitive feeling for statistical correlation does not mean that there's isn't one. You'd be very surprised at how much predictive accuracy there is in "word-choice features we can extract from a text corpus"

IBM researcher can produce pretty graph from 200 of your tweets

This seems like it would be useful for advertising, but it probably won't be. It's still difficult to link a social profile to a particlar web page view. Google and Facebook have a lot of information about everyone but they still have trouble linking it up all around the web.

I think the real value is in making predictions. Many of the things that really matter to businesses, like the stock market, government and judicial system depend heavily on human personalities. Having a better guess at what a CEO will do is valuable. Predicting what the Fed will do is even more so. There are countless boards, commitees and commissions that make influential decisions every day. Having just a slightly better idea about their actions will be based on their personalities can make a huge difference.

As the article points out, the biggest problem is that there is usually a gap between what people tweet and what they really think. The author's twitter feed was more about his persona as an online journalist than his own personal life. Many people communicate socially as who they want to be rather than who they actually are. Google could probably find out more about someone's true personality from the pages that they visit rather than the words that they use.

This technology still has great promise. One thing algorithms aren't good at is understanding the quirks of human behavior. Anything that addresses that problem moves us closer to an automated world.

> the biggest problem is that there is usually a gap between what people tweet and what they really think

I think that's a great feature, not a bug. Thank God there are aspects to our personalities that algorithm cannot predict based on just a few hundred tweets or posts.

I doubt this would be used for analysing personalities of key decisionmakers. It's applications lie in the possibility of automated (if crude) analysis of millions of people. For targeted analysis used to aid high-impact decisions, analysis done by humans will always (until we reach human-level AI) be a better tool.

Mmmmm, this could very well be just a new application of the Forer effect http://en.wikipedia.org/wiki/Forer_effect

I was going to say that what they are doing sounds like astrology or hand writing analysis. I didn't know that it was called the Forer effect, thank you. Is there even a way for them to test to see if that is happening? How do you verify personality? However, I am sure they will make tons of money with this..

...as long as the categories are narrowed to dipshit, narcissist, fundie, and smug pedant.

No they said twitter not HN. Hmm. Maybe this might be more applicable technology than just Twitter. Kind of like in TV-world the only smartphone is the iPhone maybe in TV-world the only online discussion area is officially twitter.

Oh and edited to add that the use of twitter, as far as I know, is a power law distribution where the higher the usage, the smaller the number of people, much like the distribution of internet bandwidth use over subscribers. So requiring 200 tweets means you've already split into a tiny fraction of humanity, so claiming you can analyze a tiny fraction of humanity does not equal analyzing humanity in general.

Couldn't find a link to try it out in the article. Anyone got a link?

Also worth reading is this VentureBeat interview with Michelle Zhou about her research:

http://venturebeat.com/2013/10/11/how-ibms-michelle-zhou-fig...

She needs to analyze the sayings of Lord John Whorfin:

* History is-a made at night. * Character is what you are in the dark. * Laugh while you can, monkey-boy! * Home... home is where you wear your hat.

I'm sure it would be as illuminating as her other analyses.

Is this "personality" model at all predictive, or is this just astrology?

WTF is an "IBM researcher" ? Why are we coupling ourselves to corporate entities in this day & age ? Just use her name & title. The research itself is pretty iffy.

I'll give you a serious response, as explained to me in a recent interview I had for a research scientist position: branding. Some researchers in industry are given fairly wide latitude in the topics that they are allowed to study. In these cases, the company usually gets little direct benefit from the output of the "pet" research idea. For such researchers, their indirect value is the publications and resulting press that hopefully increases the brand of the corporation in a more general sense.

I think it's a fair trade off to have to make if you're considering industrial research. I'd also note that academia isn't much different -- most of those releases also start with something like "MIT researchers..."

is this a serious question? A corporate backed researcher is one that has the financial and IT resources at their disposal to research with and use that research in the shareholder interests of that corporation. The IP does not belong to an individual, but rather to the corporation, hence the lack of attribution to the researcher.

It's like saying Sun created Java, rather than Gosling created java, both are 'correct', but the former more so.

Surely people who already have over 200 tweets, must have certain character similarities, e.g. a bit of an ego, assuming people want to know the crap they tweet.

no online demo :(

I was disappointed, too

The online demo would only need one rule:

if (true) { return "Narcissist" }

That is great because I have written less than 200 tweets, and so have most of my friends.

"Machete don't tweet"

I would love to see how my Facebook statuses portray my personality, especially were it segmented by year.

Not impressed. I can guess your personality from one of your tweets.

IBM researcher can decipher your personality from 200 of your tweets

memo to self, always keep the total number of avail tweets at 199, or 198 just to be sure.

Google and Facebook might, simply because they know everything about you and sites you visit.

Reply on news.ycombinator.com