Exploring StackOverflow data

on YouTube ↗55m

" summary

Whole StackOverflow XML dump (~100–135 GB) is preprocessed into a ~1.5 GB CSV and analyzed with CSV type provider in F#; a weekend index compares weekend vs weekday questions and finds Python skews to weekends while C# skews to workdays. Geocoding of ~5M users yields ~800k locatable entries (83% matchable) and a programmers-per-million metric by country. Community structure uses t-SNE embedding on 800 tags × 1,600 power users; text analysis uses word2vec analogies. Regression on 18M answers shows favorites and accepted answers strongly predict score, with high-leverage outliers from John Skeet.

AI-written summary · about

" chapters (6)

1. 0:00Start
2. 4:20QUESTIONS
3. 15:38QUESTION: WHERE?
4. 31:37TAGS define RELATIONS
5. 45:03STACKOVERFLOW A MERITOCRACY?
6. 49:58SIDE OF THINGS

" related talks

• 31mA Recent History of Python Through Stack Overflow Questions and AnswersSam Hames · PyCon AU 2023
• 01h 02mAnalyzing StackExchange with Azure Data LakeTom Kerkhove · NDC Minnesota 2018
• 01h 04mWhat can we learn from 750 billion GitHub events and 42 TB of codeFelipe Hoffa · Voxxed Days Singapore 2017
• 01h 03mWhat can we learn from 750 billion GitHub events and 42 TB of codeFelipe Hoffa · NDC Sydney 2017
• 25mAuto-Detection of Copied Code from StackOverflowAndrew Walker · DevConf.CZ 2020
• 19mOSPO Data PyramidCali Dolfi · Open Source Summit Europe 2025
• 36mA developer's guide to probabilistic programmingEvelina Gabasova · Lambda World Seattle 2018
• 42mBeyond bus-factor: data-visuals & algorithms to better understand Apache communitiesMarton Elek · ApacheCon@Home 2021