Social Networks Software at the Command Line
Benjamin E. Lind
9 February 2015
There is No Panacea
Software is a social artifact
Academic developers write software for their own needs as a researcher
It's a mistake to assume that the developers' needs match your own
If a software package does not exactly meet your needs, you will need to either use an additional package or write your own functions
What are Your Needs?
- Features
- Social Networks or Network Science?
- Simple Graphs or "Not-Simple Graphs?"
- Metrics, Simulation, Modeling, or Visualization?
- How to Construct a Network?
- Conventional or Flexible?
- Speed and Network Size
- Practical or Educational?
- Do you want to learn or just get it done?
Platforms
- R
- Python
- Julia
- C++
Unless you really like C++ and dislike the above languages, I wouldn't recommend pure C++ for social network analysis to anyone in the room
Package Families
Grouped by Some Common Theme
- sna
- statnet
- igraph
- tnet
- RSiena
- NetSim
- RNetLogo
- Visualization
- d3Network
- rgexf
- ndtv
- Everything else
Metrics
- sna
- Many standard and classic social network metrics
- Hard to find many features elsewhere
- igraph
- Standard metrics and those in network science
- tnet
- Two-mode
- Weighted
- Dynamic
Modeling
- statnet
- ERGM, STERGM, BERGM
- Relational event modeling
- Degree distribution, latent clusters, motifs...
- RSiena
- Network and behavioral evolution
- sna
- Blockmodels
- Bias networks
- GLM, QAP, CUG
- igraph
- Blockmodels, community detection
- Degree distribution
- Hierarchical structures
Simulations
- statnet
- ERGM estimates to simulated networks
- Permutation and QAP
- igraph
- "Games"
- Rewiring
- sna
- Erdős–Rényi-style random networks
- tnet
- Permutations
- Rewiring
- Agent-based models
- NetSim (RSiena tradition)
- RNetLogo
Visualizations
- Within R for typical needs
- sna
- network (in statnet)
- igraph
- Dynamic
- networkDynamic and ndtv (in statnet)
- d3Network for D3.js
- rgexf to export to Gephi
Network Construction
- Most rigid
- sna (simple is best)
- tnet
- Somewhat flexible
- network
- igraph
- networkDynamic
- Relatively flexible
- NetSim
- RSiena
- Very flexible
- None
Speed
- Not so fast
- sna
- tnet
- Somewhat fast
- statnet
- Very fast
- igraph
Educational vs. Practicality
- Good for education
- sna
- tnet
- Good for practical needs
- Within network packages in R, practicality is usually proporational to the research problem
Modules
- networkX
- igraph
- networkit
- graph-tool
- snap
- as well as some agent-based modeling modules
The Modules are Roughly Comparable
- Compared to R
- All come from a network science rather than social network background
- None have modeling flexibility like statnet, Rsiena
- Exceptions to comparability within Python
- networkX
- Many (most?) features, e.g., 2-mode analysis
- Written only in Python, so easy to read code
- Very slow, the others are very fast
- snap has opaque naming conventions
- graph-tool is limited in terms of simulations
- networkX
- In general, they're all quite serviceable
Packages
- Graphs
- GraphCentrality
- GraphLayout
- TikzGraphs
- RandomGraphs
- LightGraphs
- SimpleGraphs
Python modules are easily accessible
R code can be accessed as well
Benefits
- Educational rather than practical
- It will improve your programming skills
- Must think about types and network construction
- Limited existing functionality
- Must write programs when needed
- Requires you understand the process
- Decent speed
- Community
Libraries
- Boost Graph
- igraph
- NetworKit
- snap