Organizing data into taxonomies is a way to help users navigate large collections of information. Some of these collections are of relatively narrow interest and are used by trained specialists; others are for lay users to explore and take advantage of.

This paper investigates the best manner to build hierarchical indexes automatically for the consumption of data collections by lay users. Its main contributions are in (1) applying a clear focus toward addressing the need of a non-specialist, untrained base of users (much of the previous work was aimed at the needs of specialists); (2) identifying through experimental trials several metrics useful in general for ranking competing algorithmic approaches; and (3) proposing novel criteria and algorithms that appear to perform statistically better than what was tried in the past.

A shortcoming of this work is perhaps in the choice of the experimental user base: Test subjects were obtained from those willing to participate (and qualified, according to reasonable criteria) among all users with an email address on a university campus. While this is much more open than a cloister of specialists, it still contains a strong bias for minds trained--even among administrators--along the principles of classical reasoning. Such a shortcoming does not invalidate the results, but it imposes non-trivial boundaries on the meaning of “lay user.” Under this constraint, the paper candidly describes what worked well, and not so well, when creatively investigating and ranking how proficient an algorithm can be in arranging raw information in a hierarchical index most useful for a person to navigate. The result offers clearly identified outcomes (even if understandably short of perfection) and sensible recommendations for future work.