I am working on a (somewhat exploratory) clustering of linguistic attributes for data we are working with. In an experiment, we designed prompts to elicit responses, hoping to prompt some sarcasm use. We now have our sarcastic responses tagged for 10 binary (present/absent) attributes, such as the presence of extreme adjectives or use of a rhetorical question. I am trying to cluster these data (initially was for class project, but would like to clean up for better quality analysis) to find patterns of responses (e.g., direct sarcastic messages with hyperbole and extreme adjectives). I am aware of the limitations of clustering, but am using it as an exploratory analysis to guide future studies.
At the suggestion of a professor I have worked with before, I have first converted my data into three Similarity Matrices (for comparison), using the SPSS PROXIMITIES functions. Specifically, I have these same data in a matrix using Jaccard, Dice, and Russell & Rao similarity measures. While these data now "appear" continuous--each case's similarity to other cases does vary--these are still based on binary data. As such, I am unsure which method (e.g., between groups, within groups, nearest neighbor, Ward) is most appropriate to use. Additionally, beyond the method, I am unsure which measure is appropriate.
The default seems to be "Between Groups Linkage" with "Euclidean Distance." However, I do not believe Euclidean is necessarily appropriate for these originally-binary data. However, I have never worked with similarity matrices, so maybe there is no problem.
In short, I would appreciate any suggestions on how to cluster binary data transformed into a similarity matrix and if there are any concerns to be aware of. Initially, I just used "Between Groups Linkage" with "Squared Euclidean Distance" and the results are somewhat sensible, but I am wondering if this is appropriate.