I analysed about 95,000 messages exchanged with my ex‑partner. Each message was tokenised, emojis were mapped to descriptive words, and sentiment was scored with the AFINN lexicon (which assigns integers from −5 = very negative to +5 = very positive to English words). Daily mean scores were then smoothed with a seven‑day rolling average. The resulting plot tracks how our aggregate emotional tone changed over time, highlighting two breakup periods and the brief reunion between them.
DrTommyNotMD on
For an ex, you two talk too much.
NuancedFlow on
Which is you? Who initiated the breakup? It seems GF is generally more negative than BF. Do you have any pre-dating history? Could be interesting to include.
KindaMoi on
That’s very interesting. It would be great if you could share the code!
KingMonkOfNarnia on
This is insane in a somewhat neutral way and I hope you can find an equally eccentric, equally passionate statistician to eventually call a soulmate
disappointed_darwin on
This might be the most “Reddit” post of all time.
wyseguy7 on
You might consider adding volume of text messages exchanged on a secondary axis. Nice work!
acheta200 on
Seems that emotion was out-of-sync for long periods of the time. I wonder if it just noise or if there is smth to it, like one the persons being sarcastic when another is venting.
samas69420 on
maybe using afinn score isn’t the best alternative, with that kind of score sentences like “this is so fucking good” would be evaluated as negative even if they are actually highly positive, now that we have LLMs you could perform a much better analysis by using them (like you could pass each message to them and ask for a evaluation or even entire conversations) or you could also use smaller but specialized models trained only for the sentiment analysis task
BigCliff911 on
The data is skewed because you determined the tone, not an impartial opinion.
literroy on
I’m fascinated by so many things about this. First, that you got back together when the mood rating was at near its lowest. Second, how the mood rating improved (a little) immediately after the final breakup. Although, I guess that’s partly because you’re averaging your tone with hers. Looks like your tone was positive enough to cancel out her quite negative tone.
Anyway, thanks for giving us this glimpse into your life!
RespekKnuckles on
Your positiveness is notable post-breakups. Also, interesting to see your upticks shortly followed by your SO’s upticks in mood. You have (had) an effect on them.
Good data, good viz, and good follow ups in comments. Well done
andersonb47 on
This has Nathan Fielder written all over it
TAT3ST0N3 on
It looks like one of you love bombed the other in the beginning, pulled back after 3 months to gage a reaction and then breadcrumbed lower highs and higher lows of attention and validation until ultimately becoming indifferent. Lol or am I way off with my reading?
![[OC] Daily Emotional Tone in Text Messages Exchanged with My Ex (Text Analysis)](https://www.byteseu.com/wp-content/uploads/2025/04/flvvp1mgjmxe1-1536x922.png "[OC] Daily Emotional Tone in Text Messages Exchanged with My Ex (Text Analysis)")
16 Comments
**Data source**
– Personal WhatsApp chat export of ~95,000 one‑to‑one messages (BF ↔ GF), April 2023 – April 2025
**Tools**
– R
– tidyverse
– dplyr
– lubridate
– tidytext
– textdata
– stringr
– stringi
– tm
– quanteda
– quanteda.textstats
– syuzhet
– readr
– scales
– ggthemes
– zoo
– ggplot2
**Method**
I analysed about 95,000 messages exchanged with my ex‑partner. Each message was tokenised, emojis were mapped to descriptive words, and sentiment was scored with the AFINN lexicon (which assigns integers from −5 = very negative to +5 = very positive to English words). Daily mean scores were then smoothed with a seven‑day rolling average. The resulting plot tracks how our aggregate emotional tone changed over time, highlighting two breakup periods and the brief reunion between them.
For an ex, you two talk too much.
Which is you? Who initiated the breakup? It seems GF is generally more negative than BF. Do you have any pre-dating history? Could be interesting to include.
That’s very interesting. It would be great if you could share the code!
This is insane in a somewhat neutral way and I hope you can find an equally eccentric, equally passionate statistician to eventually call a soulmate
This might be the most “Reddit” post of all time.
You might consider adding volume of text messages exchanged on a secondary axis. Nice work!
Seems that emotion was out-of-sync for long periods of the time. I wonder if it just noise or if there is smth to it, like one the persons being sarcastic when another is venting.
maybe using afinn score isn’t the best alternative, with that kind of score sentences like “this is so fucking good” would be evaluated as negative even if they are actually highly positive, now that we have LLMs you could perform a much better analysis by using them (like you could pass each message to them and ask for a evaluation or even entire conversations) or you could also use smaller but specialized models trained only for the sentiment analysis task
The data is skewed because you determined the tone, not an impartial opinion.
I’m fascinated by so many things about this. First, that you got back together when the mood rating was at near its lowest. Second, how the mood rating improved (a little) immediately after the final breakup. Although, I guess that’s partly because you’re averaging your tone with hers. Looks like your tone was positive enough to cancel out her quite negative tone.
Anyway, thanks for giving us this glimpse into your life!
Your positiveness is notable post-breakups. Also, interesting to see your upticks shortly followed by your SO’s upticks in mood. You have (had) an effect on them.
what kind of sick fuck would do such a thing
https://preview.redd.it/r5ly9k83doxe1.png?width=2379&format=png&auto=webp&s=455ff5c9557920af314150b66843fc2dcecb2aba
Good data, good viz, and good follow ups in comments. Well done
This has Nathan Fielder written all over it
It looks like one of you love bombed the other in the beginning, pulled back after 3 months to gage a reaction and then breadcrumbed lower highs and higher lows of attention and validation until ultimately becoming indifferent. Lol or am I way off with my reading?