Software Collaboration Networks - Complex Networks - Lecture Slides, Slides of Data Communication Systems and Computer Networks

Download Software Collaboration Networks - Complex Networks - Lecture Slides and more Slides Data Communication Systems and Computer Networks in PDF only on Docsity!

Software Collaboration Networks

Overview

• Introduction
• Background
  • Changes
• Methodology  Data Collection  Network Topologies  Measures  Tools Conclusion
• Questions

Changes

• The addition of Github to the study
• Contains some of the same attributes to allow for a comparison
• Other communities were looked at, but they either were not large enough or did not provide enough public data.

Data Collection

• Crawling the websites using a simple Perl script and regular expressions
• Collect a project list from Sourceforge
• www.sourceforge.net/projects/projectTitle
• No specified request limit
• Check for duplicates

Github Crawling

• Using the Github API provides our data
• Limited to 60 API calls per minute
• Use multiple computers to collect all 1. million projects

Github Project Page

Developer/Project Network

Project-Developer Network

Degree

• Average number of projects worked on by a developer
• Average number of collaborations
• Average number of developers on a project

Clustering Coeficient

• Examine how likely developers are to stick together in groups
• Examine both average clustering coefficient for the entire network and the local clustering coefficient for nodes of interest

Power Law

• Previous studies have found that the Sourceforge community does follow the power law
• No such study has been done on the Github community
• Fewer developers should be apart of many project while many developers should be involved with only one project

Small World Phenomenon

• Previous studies have shown the Sourceforge community does exhibit small world properties
• Once again, no study has been done on the Github community
• Using Pajek, I will create a random network of the same nodes and edges
• Then, compare the clustering coefficient and the average shortest path

Conclusion

• Through the use of network analysis, we hope to gain a better understanding of the developers of Sourceforge and Github communities.