Automating Creation of Hierarchical Faceted Metadata Structures

Proceedings of NAACL HLT 2007, pages 244–251, Rochester, NY, April 2007. c⃝2007 Association for Computational Linguistics Automating Creation of Hierarchical Faceted Metadata Structures Emilia Stoica and Marti A. Hearst School of Information UC Berkeley, Berkeley, CA estoica,hearst@ischool.berkeley.edu Megan Richardson Department of Mathematical Sciences NMSU, Las Cruces, NM merichar@nmsu.edu Abstract We describe Castanet, an algorithm for auto- matically generating hierarchical faceted meta- data from textual descriptions of items, to be in- corporated into browsing and navigation inter- faces for large information collections. From an existing lexical database (such as WordNet), Castanet carves out a structure that reflects the contents of the target information collec- tion; moderate manual modifications improve the outcome. The algorithm is simple yet ef- fective: a study conducted with 34 information architects finds that Castanet achieves higher quality results than other automated category creation algorithms, and 85% of the study par- ticipants said they would like to use the system for their work. 1 Introduction It is becoming widely accepted that the standard search interface, consisting of a query box and a list of retrieved items, is inadequate for navigation and exploration in large information collections such as online catalogs, dig- ital libraries, and museum image collections. Instead, user interfaces which organize and group retrieval results have been shown to be helpful for and preferred by users over the straight results-list model when engaged in ex- ploratory tasks (Yee et al., 2003; Pratt et al., 1999; Kaki, 2005). In particular, a representation known as hierarchi- cal faceted metadata is gaining great traction within the information architecture and enterprise search communi- ties (Yee et al., 2003; Weinberger, 2005). A considerable impediment to the wider adoption of collection navigation via metadata in general, and hierar- chical faceted metadata in particular, is the need to cre- ate the metadata hierarchies and assign the appropriate category labels to the information items. Usually, meta- data category structures are manually created by infor- mation architects (Rosenfeld and Morville, 2002). While manually created metadata is considered of high qual- ity, it is costly in terms of time and effort to produce, which makes it difficult to scale and keep up with the vast amounts of new content being produced. In this paper, we describe Castanet, an algorithm that makes considerable progress in automating faceted meta- data creation. Castanet creates domain-specific overlays on top of a large general-purpose lexical database, pro- ducing surprisingly good results in a matter of minutes for a wide range of subject matter. In the next section we elaborate on the notion of hier- archical faceted metadata and show how it can be used in interfaces for navigation of information collections. Sec- tion 3 describes other algorithms for inducing category structure from textual descriptions. Section 4 describes the Castanet algorithm, Section 5 describes the results of an evaluation with information architects, and Section 6 draws conclusions and discusses future work. 2 Hierarchical Faceted Metadata A hierarchical faceted metadata system (HFC) creates a set of category hierarchies, each of which corresponds to a …

Content truncated to 500 words