Chapter 15. Behavioural profiles, distance metrics and cluster analysis
This chapter presents the Behavioural Profiles approach, which involves the comparison of contextual features of words or constructions in a corpus. The chapter also discusses several clustering algorithms, which are based on different distance metrics. Cluster analysis is a family of techniques that can help you discover groups of similar objects in the data. Several popular methods of cluster validation and diagnostics are discussed, which involve the computation of average silhouette widths and multiscale bootstrap resampling. The chapter also demonstrates how to interpret clusters with the help of the snake plot and effect size measures. In addition, you will learn to create and interpret scree plots, which are useful for determining the optimal number of clusters.