Monitoring a TV streaming service with AI - from PageRank to graph convolutions
summary
Liberty Global monitors a 4K streaming platform serving over a million customers using a telemetry streaming stack, Spark to scale Facebook Prophet for anomaly scores, and graph methods from PageRank to graph convolutions that propagate node features via adjacency-based message passing. A Keras-style graph layer plus node embeddings and sample weights handle heterogeneity; temporal GRU stacks address time series. Early results show 100–200 quality labels jump-start semi‑supervised root‑cause classification and reduce time spent troubleshooting.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:10Project overview Liberty Global
- 3:17TV streaming content flow
- 6:26Data sources troubleshooting
- 7:51AI ops monitoring goals
- 9:13Troubleshooting as graph problem
- 11:01PageRank anomaly detection
- 14:38PageRank limitations
- 16:46Graph convolutions deep learning
- 19:55Heterogeneous node features
- 21:41Training models sparse labels
- 23:23Feedback expert engineers
- 25:30Dynamic graphs temporal features
- 26:35Wrap-up future graph DL
- 29:11Q&A
related talks
45mReal-time, Streaming Advanced Analytics....
16mDeep Stream Dynamic Graph Analytics with Grapharis
18mEhtsham Elahi, Senior Research Engineer, Netflix @ MLconf SEA
01h 47mRadanalytics.io streaming applications workshop
42mBeyond Vectors: Evolving GenAI through Transformative Tools and Methods
46mEvent Streaming and its Potential for the GenAI World
41mAdding Generative AI to Real Time Streaming Pipelines
26mLarge Scale Graph Processing & Machine Learning Algorithms