Do you think you're cool? Perhaps, you’d say, that coolness is subjective and can’t be quantified by some kind of statistical algorithm. Well... We just did it - how cool is that?
If you Google what being ‘cool’ means, you will find Collins English dictionary’s definition: it means that someone is fashionable and attractive. But, we all know that this is not exactly how we define coolness - and that a dictionary is probably the last place a cool person would look. In fact, many people say that if we could define what cool is - it wouldn’t be cool anymore.
However, at Kubik, we love a good challenge. This is why in one of our latest projects we decided to use the magic of machine learning to try and get a grasp of what makes someone cool on social media - and why. To do this, we compared the tweets of six ‘cool people’ with those of six randomly chosen people. To get the ground truth about who is cool and who is not, we relied on a magazine article that did the work for us and identified six cool people to follow on Twitter.
How did we actually do it?
We streamed around 200 tweets from each of the cool people and each of the randomly chosen people from Twitter. Therefore, we ended up with a random sample of around a thousand cool and a thousand random tweets. In order to detect what makes someone’s text cool on Twitter, we developed a supervised machine learning model to predict the cool tweets. To define the features, we turned to academic sources and defined fourteen linguistic features that we hypothesized would be useful and that were previously found in computer science journals as valuable in distinguishing between two bodies of text.
Our resulting Random Forest machine learning model was able to predict the cool tweets with the prediction accuracy of 75%!
Based on this great result we were able to use SHAP values (SHapley Additive exPlanations), a method based on cooperative game theory. SHAP enables us to go above the traditionally used prediction task to interpretation - rather than being able to just predict cool tweets, we can also understand which linguistic features are responsible for this difference between the groups. In other words, we can finally answer the age old question - what makes someone cool. And the following were our results.
What makes tweets - cool?
Our results show that cool people write tweets:
Using fewer capital letters;
Using more first person words;
Using less punctuation;
They have shorter tweets;
Less emotional intensity;
More personal opinions and
More swear words.
Coolness, through the lens of psychology
Here is our (psychological) interpretation of the results.
Shorter tweets, less punctuation, fewer capital letters, fewer hashtags - are all about effortlessness and not trying too hard (as it takes more effort to write with a capital letter). It is also about autonomy / rebelliousness as cool people are subtly rejecting the grammar conventions. They are signalling their text is not finely crafted - and they don't care how well the tweet does.
More first person words, more personal opinions - represent autonomy (doing their own thing) and leadership. Cool people put value on themselves and are not afraid to talk about their experiences online. It is also interesting to note that many leaders who express similar patterns in writing tend to have narcissistic tendencies.
Less emotional intensity - this is literally what being cool means: "moderately cold, neither warm nor very cold", which was initially used to describe temperature and then also 'cold people'. However, in time 'cool' started also being about having autonomy over your own emotions - not letting other people work you up. This is why cool people don't express any extreme emotions on social media, they are too cool to care.
Swearing - Now, we get to the interesting part, because there are two types of coolness: popularity and rebelliousness (fitting in and standing out). In this case, swearing in tweets is about rebelliousness and autonomy. Cool people are flaunting social etiquette - they are being aware of it, but breaking it anyway. This is also typical for Twitter (or any other social media) where people express themselves more freely than in reality, because they can 'hide behind their words', so they care less about fitting in.
And there you have it. So how cool are you? Since you are the Kubik Intelligence follower, with the prediction accuracy of 100%, we say, you are very cool. If anyone says differently, just tell them to talk to us, here.