This is Not Me: Machines Evaluate My Facebook Posts

This is the beginning of an abstract look at the way my Facebook posts might be interpreted by any number of services, algorithms, job hunting boards, or even a friend from the past 10 years if certain tools are used. It’s not accurate.



I downloaded my comment history from Facebook in json format.

I hacked together this python script to do the following:

  • Remove posts to groups. I didn’t want any of my posts to groups to show up, because that feels more private and some of the groups I belong to are secret
  • Remove posts to timelines. I didn’t want any of my posts to other people’s timelines to show up because those felt like personal direct messages before messenger existed.
  • Remove birthday posts. Because birthday posts are mostly all alike and I had to remove the joy bias. I love wishing people a happy birthday but this is the highest possible level of happiness and honestly when I wish someone a happy birthday on facebook I do it because I care and I want them to feel loved but the JOY I feel is maybe … 1 second long?

So after cleaning up my posts, I pasted my posts into Watson’s Tone Analyzer. I got a whole lot of sentiment but I don’t agree with a lot of it.

Watson wrong

In one of the more horrifying examples of Watson misinterpretation, I made the following comment on February 21 2018 after the shooting at Marjory Stoneman Douglas High School in Parkland,Florida:

Facebook post:

line1: “The son of a friend of a friend was killed in the Florida shooting.”

line2 : “Please consider donating in his memory towards something positive.”

“score”: 0.587877, “tone_id”: “joy”, “tone_name”: “Joy”

“score”: 0.901894, “tone_id”: “analytical”, “tone_name”: “Analytical”

—JOY?!— Stripped of all context I can only assume that the word positive was enough to trigger the analysis. But it’s so, so wrong.

Amazon Comprehend Wrong

Facebook Post: ​I just read about ghosting. I’m kind of horrified — even my first boyfriend at 14told me the line “it’s not you it’s me” to my face. Have any of you experienced this — if not with a dating partner, a friend? relative?

{ “Sentiment”: “NEUTRAL”, “SentimentScore”: { “Mixed”: 0.06482646614313126, “Negative”: 0.3775499761104584, “Neutral”: 0.493772953748703, “Positive”: 0.06385056674480438 } }

I don’t feel neutral about this topic. I definitely feel negative. But, the machine says I’m neutral on it.

I really don’t want to leave Facebook.

  • I lived through MySpace, Friendster, Google+ and it would be time-consuming and likely no easy to port it somewhere else.

  • An email lists updating my network with my random thoughts or questions or photos isn’t a good alternative.

  • I have 10 years of connection-building on the site that I can’t easily move.

I’m going to Paris this week for the first time, and asked my network for suggestions on where to go. I could spend hours on a corporate travel website forum, google a blog, stare at a seasoned thread on where to go on reddit, but that wouldn’t make the trip feel like my own. I wanted personal stories and thoughts from people I know and care about. I got what I wanted. I plugged places into google maps and have a very good general idea of where to go, and a way to connect with my friends and family when I get back (“I went to that bakery you suggested!”)

Comments on facebook have been one of the greatest sources of community in my feed. I get joy out of having people I care about offer suggestions and advice. In the original facebook data dump from a month ago I was fascinated that comments weren’t included. This made the data dump feel like a diary. Since the Cambridge Analytica scandal I can now see comments with the new data. That is maybe for another day.

For what it’s worth, my post about going to Paris brought me a whole lot of sustained joy (maybe an hour’s worth!) greatly due to feedback I received. But it only rated a .75 on the joy scale compared to the .98 birthday joy. That too, is just not me.

“text”: “I’m going to Paris in May! [rest of post private].”, “tones”: [ { “score”: 0.753788, “tone_id”: “joy”, “tone_name”: “Joy” }