KEPC-Push: A Knowledge-Enhanced Proactive Content Push Strategy for...
summary
Proposes KEPC-Push, which uses knowledge graph embeddings (TransE, 128‑D vectors) plus collaborative‑filtering/Skip‑gram to infer popularity for fresh videos, then runs region‑level allocation with CH‑boost forecasting, K‑means clustering and a 5‑minute decision cycle to pick replica counts and edge placements. Trace‑driven simulation on 16M requests shows ~20% peak CDN bandwidth savings, ~7% average download speed improvement and ~8% fewer cross‑region requests, while keeping replica churn and peak node overloads lower than baselines.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:52Video feed streaming challenges
- 4:46KEPC-Push overview
- 6:55File feature embedding
- 9:06Popularity correlation graph
- 11:59File deployment scheduling
- 14:52Experimental setup
- 15:56Performance improvements
- 18:52Ablation study
- 19:34Summary
related talks
20mDataEngBytes 2023 - PER-T1-06 - Kamal Hossain
06mWalking Around Your Nearest Neighbors
25mDeliver viewing experiences for super fans with Amazon Personalize
26mBeyond the Vector DB
20mLayering every technique in RAG, one query at a time
11mBuilding Real-Time Feature Pipelines
32mBuilding an recommendation system for IPTV
19mX1 Recommendations Platform: Started from the bottom..