Natural language processing (NLP) is a powerful tool to understanding how a major disease — in this case, obesity — is perceived on social media.

First,though, let me put the title I used in realistic human context.

“Eww, fat people disgust me!!!”

That’s a real example of an offensive tweet about obesity.

Here’s another, more offensive tweet:

ATTENTION FAT B***S: Stop wearing tight a* pants and leggings, that s*** is nasty! Wear baggy jeans or overalls…

And here’s the one tweet I found most disturbing:

@USERAME, you’re an ugly fat b****. kill yourself.

These horrible tweets are just three examples pulled out from a review of 2.2 million social media posts over sixty days, using natural language processing (NLP) techniques.

The study I am referencing can be found in the very insightful 2014 article Obesity in Social Media: a mixed methods analysis, written by the research folks at the Health Communications and Informatics Research Branch, Behavioral Research Program of the National Cancer Institute.

This study is important because it answers an important question: Are these cruel posts/tweets just aberrations?

Sadly, the answer is no.

Using NLP techniques and machine learning, and using those 2.2 million posts from Twitter, Facebook, blogs and forums, the most common linguistic bigrams were extracted for the keywords: FAT, OBESE/OBESITY, and OVERWEIGHT. (Bigrams are word pairings which help determine sentiment and context of a given keyword.)

The keyword FAT occurred over 1.2 million times, and the six most common bigrams were:

Fat a** (100,632)

Fat girl (74,168)

Fat people (61,724)

Fat so (51,233)

Fat b****** (39,897)

Fat kid (36,371)

Clearly these FAT bigrams are derogatory.

Moving on, the keywords OBESE/OBESTY occurred over an order of magnitude less in frequency — roughly 75,000 times — and had a much more clinically neutral tone:

Childhood obesity (6,348)

Obese maybe (5443)

Kids obese (5,242)

Obesity http (3,878)

Morbidly obese (3,397)

Obesity epidemic (3,217)

And finally the term OVERWEIGHT occurred even less frequently — around 26,000 times, with these common bigrams:

Overweight people (1,121)

Overweight http (789)

Only overweight (532)

Overweight thing (513)

System overweight (474)

Overweight women (449)

Here are a few general observations.

First, the term FAT is totally dominating the social media space in the public conversation about obesity, and this domiantion is by over a factor of ten.

Second, the FAT keyword-based postings and discussions are for the most part misogynistic and negative, often crossing the line into clear cyber-bullying.

Third, the commonly used keyword terms OBESE/OBESITY/OVERWEIGHT occur at a much lower frequency, but do tend to be neutral to positive.

But, apart from the insight that the social media universe can be both rude and negative (something most of us already know), what does this mean for us physicians?

The message is loud and clear: Physicians need to be more engaged on social media!

Look, I am sympathetic to the argument that most physicians find social media a waste of time; I tend to agree with this if a doctor engages in social media without a mission, and without some basic understanding of what social media can and cannot do to support their mission.

But, if we look at ourselves outside of our personal narrow career focus, and instead see ourselves as a unique part of the citizenry — the part of the citizenry which can bring vocal expertise and compassion to such a devastating medical problem as obesity — it is clear we have a responsibility to be the voice of compassion and reason in the public space.

Why?

Because if we don’t, who will? The tweeter who thinks someone should kill themselves for being “fat”?!

So, if you are a physician, read the linked article above. It’s powerful, and a must read for anyone interested in obesity, AI, and social media.

Then consider the possibility that learning how to use social media effectively may actual be an ethical responsibility, both as a physician and a citizen.