The Application MCA with K-Means to Affiliation Networks
This research examined the application of multiple correspondence analysis with k-means to affiliation network data and explores the limitations that this technique may encounter with large data sets. Previous research has shown that multiple correspondence analysis (MCA) is an appropriate technique to visualize affiliation network data. This research provides further support for the use of MCA and expands on it to understand the segmentation that exists within the network data. There are two approaches to affiliation networks: the direct and the conversion method. The conversion method splits the two-mode network into two one-mode networks. The direct method is the preferred method for affiliation networks; however, segmentation techniques result in two main segments: events and individuals. MCA was previously examined and found to provide acceptable visualizations that display both network modes; however, the inference into segmentation is purely visual. MCA with k-means allows for the entire data set to be segmented providing clusters based on both modes of the network. The technique was compared to SNA, CA, and MCA. The different techniques were compared using a well-known affiliation network sample. The results demonstrated that this technique is appropriate, provided more information than just MCA, and is more informative for understanding clusters in affiliation network analysis. The research also applied MCA and MCA with k-means to large randomly generated networks. This revealed that while both are appropriate techniques there may be limitations with MCA and MCA with k-means. Specifically, when the networks are large, the utility of the technique decreases. This decline in utility due to the difficult to interpret visualizations and the amount of inertia for which the visualizations account. When the network is visualized in two dimensions, but only accounts for a very small amount of inertia, the visualization is inaccurate. Overall this research found that MCA and MCA with k-means are appropriate techniques for affiliation networks, but their ability to accurately visualize affiliation networks may be limited when the network is large.
Smith, Matthew J, "The Application MCA with K-Means to Affiliation Networks" (2020). ETD Collection for Fordham University. AAI27737296.