Resolving Task Specification and Path Inconsistency in Taxonomy Construction

Taxonomies, such as Library of Congress Subject Headings and Open Directory Project, are widely used to support browsing-style information access in document collections. We call them browsing taxonomies. Most existing browsing taxonomies are manually constructed thus they could not easily adapt to arbitrary document collections. In this paper, we investigate both automatic and interactive techniques to derive taxonomies from scratch for arbitrary document collections. Particular, we focus on encoding user feedback in taxonomy construction process to handle task-specification rising from a given document collection. We also addresses the problem of path inconsistency due to local relation recognition in existing taxonomy construction algorithms. The user studies strongly suggest that the proposed approach successfully resolve task specification and path inconsistency in taxonomy construction.