Understanding Social Networks from the Local Behaviours within the Network

Watch the video

Military / Coalition Issue

Understanding networks is critical to multi-domain operations. Network structures in dynamic contexts are rarely static, and often function with considerable complexity. These networks could represent diverse concepts such as communication, social network activity, online media or many other issues. To understand such networks, traditional graph theory has limitations because we can often only observe part of the network. Therefore, being able to make assessments through the presence of substructures is important.

Core idea and Key Achievements

This achievement summarises the progress made in using graphlets (small induced substructures) to analyse potentially complex network structures without recourse to global network characteristics. This includes:

  • Applications to social media for the detection and assessment of disinformation, disruption and controversy;
  • Alternative ways to understand paths through the network by the way that they “cut through” graphlets, thus enabling prediction as to how information and influence is liable to spread;
  • Ways to use graphlets (through network embedding) to provide an alternative network representation, which enables machine learning of patterns and behaviours.

Implications for Defence

The military can use these techniques in settings where only partial network information is available – the presence of graphlets can be collected from dynamic, disconnected and temporal snapshots of networks and used for analysis through these techniques. Further, they avoid the need for processing the language content of social media.

Readiness and Alternative Defence Uses

The research has been carried out at a fundamental level but has been applied in interesting real-world network scenarios, including those relating to disinformation, and COVID-19. This means that the work can be readily applied to other data sets of operational relevance.

Resources and References

  • Davies, C., Ashford, J., Espinosa-Anke, L., Felmlee., D, Preece, A., Srivatsa, M., Turner, L., and Whitaker, R.M., (2021), Multi-Scale User Migration on Reddit, Workshop on Cyber Social Threats, International Conference on Web and Social Media (ICWSM), accepted for publication.
  • Hudson, L, Whitaker, R.M., Allen, S.M., Turner, L., Felmlee, D., The Centrality of Edges based on their role in Induced Triads, (2021), IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM), in submission.
  • Tu, Kun, Jian Li, Don Towsley, Dave Braines, and Liam D. Turner.
  • gl2vec: Learning feature representation using graphlets for directed networks.” In Proceedings of the 2019 IEEE/ACM international conference on advances in social networks analysis and mining, pp. 216-221. 2019.
  • Ashford, James, Liam Turner, Roger Whitaker, Alun Preece, Diane Felmlee, and Don Towsley. “Understanding the signature of controversial Wikipedia articles through motifs in editor revision networks.” In Companion Proceedings of The 2019 World Wide Web Conference, pp. 1180-1187. 2019.

Organisations

Cardiff, IBM UK, IBM US, UMass, Penn State