Identifying Communities of Trust and Confidence in the Charity and Not-for-Profit Sector: A Memetic Algorithm Approach

In this study we analyse complete networks derived from field survey and market research through proposing an efficient methodology based on proximity graphs and clustering techniques enhanced with a new community detection algorithm. The specific context is the charity and Not-For-Profit sector in Australia and consumer behaviours within this context. To investigate the performance of this methodology we conduct experiments on the network extracted from a dataset that contains responses of 1,550 individual Australians to 43 questions in a quantitative survey conducted on behalf of the Australian Charities and Not-for-Profits Commission to study the public trust and confidence in Australian charities. Here, we generate the distance matrix by computing the Spearman correlation coefficient as a similarity metric among individuals. Then, several types of k-Nearest Neighbour (kNN) graphs were calculated from the distance matrix and the new community detection algorithm detected groups of consumers by optimizing a quality function called "modularity". Comparison of obtained results with the results of the BGLL algorithm, a heuristic given by the publicly available package Gephi and the MST-kNN algorithm, a graph-based approach to compute clusters that has several applications in bioinformatics and finance, reveals that our methodology is effective in partitioning of complete graphs and detecting communities. The combined results indicate that behavioural models that investigate trust in charities may need to be aware of intrinsic differences among subgroups as revealed by our analysis.