An Agent Approach for Modularity Maximisation in Network Community Detection

Activity: SupervisionMaster Dissertation Supervision


In this project we will implement the modularity maximization algorithm that is frequently used in community detection, using an agent-based approach. In network analysis, modularity measures the strength of a community partition for networks with the consideration of the degree distribution of nodes. Given a well-defined community effect, the maximum community effect with the optimal partition can be considered as a trace minimization problem. Conventionally, process of solving the optimisation problem includes the calculation of the modularity matrix and then finding the top k eigenvectors as inputs to clustering algorithms such as k-means for community detection. The issues with the conventional method become apparent in two aspects. 1. The computational power required for the calculation grows non-linearly with the number of elements in the matrix; 2) the change of network structure will lead to the re-calculation. In this project, networks and their elements will be modelled as agents, structure of the network will be formed as interactions between agents, thus the optimal modules can eventually be formed through agent interactions while the effect of addition and removal of agents to the network structure will be gradually reflected through the dynamics of the agent interactions. The project will be implemented using the NetLogo-specific language and the evaluation will be done on the NetLogo platform.
Period14 Dec 2023
Degree of RecognitionInternational