More Then Words : Business Applications of Recurrent Networks
summary
Red Hat uses recurrent neural networks to predict churn and prioritize escalations, training on 18 months of behavior to forecast 6 months ahead and integrating web logs (Hadoop/Redshift), SQL, scikit‑learn preprocessing, SVD/LDA text features and mmap numpy storage. A gradient‑boosted model (XGBoost) labels missing feedback; RNN predicts churn 18‑month→6‑month and XGBoost for feedback labels. Results show RNNs beat linear/GBM baselines by ~10–20%, detect ~100 more at‑risk accounts, and raise escalation true positives from 15% to 20%.
AI-written summary · about
chapters (8)
- 0:00Deep learning at Red Hat
- 1:25What is deep learning
- 4:56Recurrent neural networks
- 9:10Customer asset churn prediction
- 22:56Support case escalations
- 24:39Word2Vec for baseball players
- 25:22Semantic similarity model
- 26:05Q&A
related talks
38mUsing Recurrent Neuronal Networks for Customer Predictions
31mThe Advantages of Deep Learning with Recurrent Neural Networks
18mTransforming Financial Decision Intelligence Through Advanced NNs
23mCustom-made games with machine learning and big data
20mBuilding Intelligent Retention Engines: Real-Time Churn Detection
01h 39mApplying Rigorous Machine Learning Methods in Business Strategy
47mBuilding ML practices to address the top four use cases
24mUnlocking revenue: newcomer classification for boosted profits