My conversation with my now wife Chloe on Facebook messenger is quite the compilation of data, with 41,429 messages over 3 years. We talked into the late hours, and in small moments during the day. As we spent more time actually together as our courting progressed, we messaged less and less. We trusted zucc with a great deal of our data, and I wanted to get an idea of some of the things Facebook could learn about us with its algorithms.

The times you choose to message someone during a week implies a very different conversation. Hundreds of messages sent in a few hours between dinner and bed are very different than a few messages (presumably coordinating dinner or something like that) sent at midday. Before we were dating, the majority of messaging occurred late at night.

Week-hour messaging heat map between Brent Halonen and Chloe Halonen from Oct-30-2015 to Jun-15-2016
During this time, the messaging was the ‘meat and potatoes’ of our interactions, otherwise seeing each other on the weekends at church events or in coffee shops. We started dating in the summer of 2016, and our messaging shot up across all the bands. Clearly high levels of interest
Week-hour messaging heat map between Brent Halonen and Chloe Halonen from Jun-15-2016 to Jun-15-2018
We hung out very frequently, and I am surprised there isn't a fade in messaging later in the day. When we got married, messaging outside of work totally collapsed, as I go home every day after work.
Week-hour messaging heat map between Brent Halonen and Chloe Halonen from Jun-15-2018 to Apr-15-2019
What we talked about varied over time. I ran a latent Dirichlet allocation analysis and extracted some topics, and then plotted how those topics varied over time.

plot 1 plot 2 plot 3 plot 4 plot 5 plot 6 plot 7 plot 8 You can see some interesting trends in the data. Most interesting to me is the increase in popularity of conversations around food, implying as time goes on, the larger and larger part of our conversation is dominated by food.

I've since moved my messaging to Keybase to avoid being datamined by zucc. I'd recommend moving your personal communications to encrypted channels as well. No reason to give Facebook all that power. Clearly, there is quite a bit of information to be exploited in our personal messages.

If you have a large data set you want analyzed, describe your data and what you might want out of it here.