The author's finding 'partisan taunting' was a neat study and illustrative example of using their computer-assisted clustering method. Yet, I am still uncertain as to what provides a useful clustering.
I can understand the usefulness of running multiple clustering models to find the most interesting groupings, but I fear losing the value of a question by simply grouping text in every possible combination. How can a researcher rigorously select the right groupings without cherry-picking the data that best fits his model?
The author's finding 'partisan taunting' was a neat study and illustrative example of using their computer-assisted clustering method. Yet, I am still uncertain as to what provides a useful clustering.
I can understand the usefulness of running multiple clustering models to find the most interesting groupings, but I fear losing the value of a question by simply grouping text in every possible combination. How can a researcher rigorously select the right groupings without cherry-picking the data that best fits his model?