Happiness inside a job: a social network analysis
summary
Uses two years of mobile app data from 34 companies and ~4,400 employees to predict employee turnover by combining network features and classic ML The concrete claim is that a simple gradient boosting model finds likability ratio (likes over total reactions) as the strongest predictor alongside company-level posting rates and an NMF-derived social component The repo contains the CSVs, feature pipeline, balancing tricks and a Bokeh dashboard to inspect network plots and feature importances
AI-written summary · about
chapters (23)
- 0:00Predicting employee happiness
- 1:04Social network data predicts turnover
- 2:12Data structure and info
- 3:42Employee interactions and voting
- 5:52Daily happiness trends and behavior
- 7:18Posting days impact engagement
- 8:01Employee turnover timelines
- 9:05Data visualization and cleaning
- 9:49Graph theory for social networks
- 12:39Directed graph representations
- 13:50Real-world company data analysis
- 16:05Data filtering and prediction setup
- 17:52Choosing graph methods
- 18:36Creating predictive features
- 20:24Non-negative matrix factorization
- 22:17Training gradient boosting classifier
- 23:06Key features for turnover risk
- 25:54Likability predicts retention
- 26:43Three takeaways on happiness
- 28:29Answering audience questions on analysis
- 31:49Combining graph and non-graph features
- 32:20Bokeh dashboard for visualization
- 34:45Handling biased data
related talks
37mExtracting Insight From A Muslim Marriage App
01h 26mBuilding an Interactive Network Graph to Understand Communities
01h 33mBuilding an Interactive Network Graph to Understand Communities
41mCity attractiveness seen through Twitter
55mSocial Listening for Richer Data, Unstructured Analytics
46mComputing Professional Identity for the Economic Graph
36mHunting unicorns with Network analysis
05mInterview with Alex Willcock