Complex networks sparsification: methods and tools
summary
Walks through sparsifying a NIPS paper–word bipartite network and shows concrete effects: starting graph is ~1,500 papers and ~16,000 nodes with ~1.1 million edges, cutting by weight at threshold 141 drops edges roughly tenfold to ~160k and yields a usable sparse core, while spectral comparison finds a local minimum at 151 that gives even sparser structure with little topological loss. Implements methods with NetworkX, pandas and NumPy and demonstrates a configurational model pruning that preserves the giant component while removing ~90% of incidental edges.
AI-written summary · about
chapters (10)
- 0:00Complex networks sparsification
- 2:07Networks need pruning
- 8:43Bipartite networks
- 17:22Document similarity projection
- 27:17Pruning by link weight
- 36:56Degree and strength analysis
- 44:26Spectral sparsification
- 53:06Configurational pruning models
- 1:02:13Probability-based pruning
- 1:05:28Summary
related talks
01h 12mSocial Network Analysis in Python
30mCitation Network Analysis from Scratch
24mHighlights from Current Programs part 2:
44mScalable Algorithmic Primitives for Data Science
55mDistribution-Free Models of Social and Information Networks
48mBig Graph: Big Data applied to giant, dynamic graphs
16mUsing Network Analysis to Detect Kickback Schemes Among Medical Providers
10mNetwork Structure and Privacy: The Re-Identification Risk in Graph Data