I analysed 20 years of my chats

9 hours ago

The author analyzed 20 years of digital chats (1.2M messages) to understand relationships better, inspired by Tim Urban's 'Your Life in Weeks' visual.
They attempted journaling but found it incomplete, then built a personal CRM from archived messages across platforms like ICQ, VK, Twitter, Instagram, and Telegram.
Data processing involved parsing archives, filtering noise (e.g., short fillers, emojis), and using LLMs for classification and name resolution with a structured JSON output.
Sentiment analysis was refined to include directional tags (my state, counterpart's, mutual) to capture emotional dynamics, revealing patterns like warmth declines or transactional shifts.
Key metrics included message volume, vocabulary overlap (Jaccard similarity), session patterns, response times, question rates, and endearment frequency over relationship stages.
Findings showed friendship half-lives, attrition rates (losing ~20 contacts/year without moving), and that total conversation-days remained stable despite network shrinkage.
The author discovered personal traits, like defaulting to advice over listening, and chronotype effects (evening vs. daytime messaging affecting network size).
The project filled the 'empty weeks' with meaningful interactions, revealing selective memory, and provided insights to potentially improve friendships.
Technical challenges included handling large data volumes and visualizing relationships in 3D, with the outcome being a deeper understanding of social patterns.

Hasty Briefsbeta