Date of Award
2019
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Computer Science : Modeling and Simulation
Committee Chair
Mikel D. Petty
Committee Member
Gregory Reed
Committee Member
Daniel Rochowiak
Committee Member
Allen Wilhite
Committee Member
Huaming Zhang
Subject(s)
Social sciences--Network analysis, Social networks, Personality, Genetic algorithms
Abstract
Social structures and interpersonal relationships may be represented in abstract mathematical objects known as social networks. A social network consists of nodes corresponding to people and links between pairs of nodes corresponding to relationships between those people. Social networks can be constructed by examining groups of people and identifying the relationships of interest between them. There are circumstances where such empirical social networks are unavailable, or their use would be undesirable. Consequently, methods to generate synthetic social networks that are not identical to real-world networks but have desired structural similarities to them are valuable. A process for generating synthetic social networks based on attributing human personality types to the nodes and then stochastically adding links between nodes based on the compatibility of the nodes’ personalities was developed. Four algorithms for finding an effective assignment of personality types to nodes were developed and tested. Using the Myers-Briggs Type Indicator as a model of personality types, a compatibility table used by the algorithms was created. The four algorithms were evaluated for realism as measured by the similarity of the synthetic social networks to real-world exemplar networks. Based on 20 standard quantitative network metrics, synthesized social networks were compared to 14 real-world exemplar networks. Custom implementations of two randomized algorithm classes, Monte Carlo and Genetic, produced more realistic networks than the classic Erdős-Rényi algorithm. Two new heuristic algorithms, Probability Search and Compatibility-Degree Matching, produced more realistic networks than the well-known and widely-used Configuration Model algorithm. To confirm that the algorithms’ effectiveness was independent of a specific personality type model, 15 Iterated Prisoners’ Dilemma strategies were treated as personality types. The strategies were implemented, an Iterated Prisoners’ Dilemma round-robin tournament was conducted, and the tournament’s results were used as a personality compatibility table. The new Compatibility-Degree Matching algorithm again produced more realistic synthetic social networks than the Configuration Model algorithm. Finally, a new randomized algorithm to synthesize a sequence of revised social networks representing the evolution of a social network over time was developed. A Turing test showed that the synthesized social network sequences were indistinguishable from real-world exemplar sequences of evolving social networks.
Recommended Citation
O'Neil, Daniel Anthony, "Synthesizing realistic social networks using personality compatibility" (2019). Dissertations. 178.
https://louis.uah.edu/uah-dissertations/178