Unsupervised Expressive Rules Provide Explainability and Assist Human Experts Grasping New Domains

Approaching new data can be quite deterrent; you do not know how your\ncategories of interest are realized in it, commonly, there is no labeled data\nat hand, and the performance of domain adaptation methods is unsatisfactory.\n Aiming to assist domain experts in their first steps into a new task over a\nnew corpus, we present an unsupervised approach to reveal complex rules which\ncluster the unexplored corpus by its prominent categories (or facets).\n These rules are human-readable, thus providing an important ingredient which\nhas become in short supply lately - explainability. Each rule provides an\nexplanation for the commonality of all the texts it clusters together.\n We present an extensive evaluation of the usefulness of these rules in\nidentifying target categories, as well as a user study which assesses their\ninterpretability.\n