Building a Roadmap of Human Emotion: EMOTIONS & WORDS

  • September 13, 2012
  • Blog

Hubs and Gateways: Previously, we uncovered the Big Five Human Mindsets and showed how current emotions can predict future mindsets. This week we are going to reveal how and why people transition from mindset to mindset. Specifically, key things you will take away from reading this entry will be:

  • Which emotions work as stepping stones towards a happier and more positive state of mind.
  • That the things that makes you become happy are not the same as the things that help you stay happy.

Every time you feel a given emotion, you’re coming from some previous emotion, and you’re on your way to some future emotion. Recall from last week’s post that each mindset has a set of “Sticky Emotions”. This week we will discuss other properties that emotions can have, considering both the future and the past.

We can think of emotions as transit stops. Some stops are hubs — people are coming from everywhere and departing for everywhere. In contrast, other stops are like passport stations at a border — people are coming from one general area and going to a different general area.

To capture this distinction mathematically, we defined a “Hub score” metric. An emotion with a high hub score would have a roughly equal amount of traffic going back and forth between any two different color mindsets. The smaller the difference in the amount of traffic going in each direction (say, the difference between the amount going from red to green and the amount going from green to red), the more hub-like an emotion is.

We found that the top “Hub Emotions” are tired, happy, excited, cold, hungry, cheerful, sleepy, confused, hot, and exhausted. True to form, these emotions are circumstantial, and do not tell us much about a person’s general emotional history. They are like hub airports – knowing that someone is going through that airport does not tell you much about where the person has been or where they are going.

To give an example of how uninformative a Hub Emotion is, we show the probability of experiencing any given emotion one day before and after feeling tired. As you can see, the distributions are roughly the same.

One day before feeling tired

One day before feeling tired

One day after feeling tired

One day after feeling tired

The size of these nodes is proportional to the probability of feeling each emotion one day before and after reporting feeling tired. The colors represent the cluster the emotion belongs to.

What about the emotions with the smallest Hub score? These emotions are more informative, and predict specific mindset changes.

The emotions with the smallest Hub score are: relievedirateenviousintimidatedsneakymorosestormyokayjubilant, and encouraged.

Can we also find the emotions that predict a positive mindset change? When an emotion is a stepping stone between two other emotions, we say that an emotion changes your mindset for the better if the future emotion has a higher valence than the previous emotion. We computed the average change in valence (considering a time decay) to find the positive effect of each emotion. We call the highest scoring emotions “Gateway Emotions” because they usually take you to a higher-valence mindset. Conceptually, Gateway Emotions are bridges from a negative feeling to a better one; they are the emotions you feel while you are on your way to feeling better.

The emotions that are the top gateways are: relievedokaycalmencouragedhopefulnumboptimistic, and hysterical.

Note that there is a difference between being well and getting better. In absolute terms, feeling “numb” or “okay” isn’t great, but these emotions are gateways because on average, people who felt these emotions were in a bad place previously and were on their way to somewhere better. The inclusion of hysterical in this list was a surprise to us. We think this means that people go through cathartic states of mind that catalyze necessary changes in their lives. Gateway Emotions are usually very transient, while Hub Emotions are not necessarily transient. If you are on your way to feeling better, you are probably not going to stick around and feel “relieved” for days on end.

To give an example of a typical Gateway Emotion, we show the probability of experiencing any given emotion in the two days before and after feeling relieved. Note how the day before, many feel surprised, presumably because something unfortunate came up. Also note how general depression and anxiety peak one day before and dramatically drop one day after.

Two days before feeling relieved

Two days before feeling relieved

Two days after feeling relieved

Two days after feeling relieved

One day since feeling relieved

Two days since feeling relieved

Besides Hub Emotions and Gateway Emotions, emotions can be characterized in numerous other ways. Due to space constraints, we are only commenting on the most interesting ones we found. Before moving on to the next topic, we can briefly comment on one last characteristic of emotions: the degree to which experiencing an emotion puts you back in the mindset you came from. We call the emotions that do this with the highest probability “Reflecting Emotions”. The top ten Reflecting emotions are: horny, depressed, rejected, nerdy, devastated, blessed, tearful, lonely, dirty, and delighted.


What Words Characterize Emotional Changes?

We just showed that empirical data can tell us a lot about the way humans experience sequences of emotions through time. That is, we discussed the “how” of the way emotions happen. Now, thanks to the descriptions users can put in their emotion updates, we can also learn a little about the “why” of emotions.

For example, when you see someone who is currently happy, how would you tell whether he or she was sad a few days ago? What would make his or her description different than the description of someone who has been feeling well consistently?

To find out, we analyzed the lexical signatures of the descriptions of emotions of a specific color when they transition to other colors. What makes a description special when the user had been feeling blue for the last few days but is now in a green mindset? It turns out that descriptions that fall in this category include the word better 3 times as often as compared to all other updates. (This doesn’t mean the word “better” is among the most common in this category. It means that if you take an update and it says “better”, it is more likely that it is a green update of a person who was in a blue emotion.) This was the single best predictor that the update is in that category, but there were other words and emoticons that were used around twice as much as compared to all updates: thanks, helped, :D , finally, friends, amazing, talked, EP, and wonderful.

We were happy to find that mentioning EP (the Experience Project) on a status update is evidence that the person is getting better; perhaps engaging with the site has a therapeutic effect on the users? As a comparison, while words such as therapy, antidepressant and psychoanalysis were not uncommon, they were not particularly predictive of getting better (however, they were more common in green emotion updates).

On the other hand, the words that characterize the green updates that come from a green emotion (updates that stay in the green cluster) are the following: Wonderful, blessed, amazing, beautiful, happy, God, great, loves, special and  :-) in that order.

We can see how these words are usually used in the context of a general appreciation of life itself rather than specific circumstantial events. And, more importantly, we have shown that what makes you become happy is not the same as what makes you stay happy.

Now, what are the words that predict a transition in the reverse direction? Most of these words suggest that something happened to the person that was out of his or her control, such as sickness, an accident, or the death of a friend or relative. These words are: headache, flu, miss, :( , missing, stomach, hurt, sad, hurts, throat and died.

Finally, the words that characterize blue updates that come from other blue updates suggest more permanent and less circumstantial conditions. These words are: miss, criminal, lonely, abusive, sad, lost, sick, hate, :( and alone.

These four possible transitions represent only a small subset of the 25 possible color transitions, but we will only focus on these due to a lack of space. However, there are a few interesting words that we thought were worth sharing from this investigation. For example, it turns out that the word coffee is 4.5 times more common in updates that transitioned from blue to yellow. What does this mean? This shows that if you are depressed and you drink coffee, you might not get better but you are likely to be turned on (other words that predict this transition from blue to yellow are shower, caffeine and temperature).

In addition, from orange to green we have: :D , Friday, done, yay, finally, :) , nice, wonderful, great, beautiful, weekend and finished. As you can see, many of these words describe the end of circumstances and temporal intervals. In these words tens of thousands of stories are being condensed, from a girl that was bored all week and is happy it is finally Friday, to a man who has finished a project and is now done.

We also found that the list of words that predict a future transition to a green emotion from a red emotion includes words that evoke uncertainty: waiting, worried and wondering. These words describe situations that cause anxiety in the present but are likely to be met with relief in the next few days.

In future articles,  we will discuss how age and gender affect your emotional experience.


The Kanjoya Team

Graphs generated with Gephi.


Leave a reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>