A Support Tool for Deriving Domain Taxonomies from Wikipedia

Proceedings of Recent Advances in Natural Language Processing, pages 503–508, Hissar, Bulgaria, 12-14 September 2011. A Support Tool for Deriving Domain Taxonomies from Wikipedia Lili Kotlerman∗, Zemer Avital§, Ido Dagan∗, Amnon Lotan♯, Ofer Weintraub§ ∗Bar-Ilan University, §Orca Interactive Ltd., ♯Tel Aviv University lili.dav@gmail.com, zemer.avital@orcainteractive.com, dagan@cs.biu.ac.il, amnonlot@post.tau.ac.il, ofer.weintraub@orcainteractive.com Abstract Organizing data into category hierarchies (taxonomies) is useful for content discov- ery, search, exploration and analysis. In in- dustrial settings targeted taxonomies for spe- cific domains are mostly created manually, typically by domain experts, which is time consuming and requires a high level of ex- pertise. This paper presents an algorithm and an implemented interactive system for automatically generating target-domain tax- onomies based on the Wikipedia Category Hierarchy. The system also enables human post-editing, facilitated by intelligent assis- tance. 1 Introduction Hierarchies of category names (taxonomies) are very useful for effective information access (K¨aki (2005), Stoica et al. (2007)). When geared for a specific domain or data collection, such hierarchies can highly benefit the tasks of content discovery, search, exploration and analysis. Our project, car- ried out by the Natural Language Processing group at Bar-Ilan University and Orca Interactive Ltd., aimed at semi-automatic generation of a taxonomy for the domain of general video content in order to enhance search and improve recommendations in a personalized video recommendation system. This paper delivers two main contributions: (1) a novel algorithm for automatic generation of target-domain taxonomies and (2) an interactive taxonomy editing tool, which helps human editor to post-edit and improve automatically generated taxonomies by providing her with intelligent as- sistance. Automatic taxonomy generation approaches can be roughly divided into two classes: corpus-based and knowledge-based. We suggest a knowledge- based algorithm, deriving focused target-domain taxonomies from the Wikipedia Category Hierar- chy (WCH). WCH covers a very wide range of top- ics and is assumed to embed smaller taxonomies suitable for specific domains. The algorithm is thus aimed at extracting such taxonomies from WCH. Since automatic techniques for taxonomy cre- ation are not accurate enough, in real-life appli- cations some human post-editing is usually em- ployed. Our taxonomy editing tool was designed to facilitate this process. It provides the editor with intelligent assistance, based on statistical similar- ity in a domain corpus along with WCH. Our ini- tial experiments in the video domain show consid- erable reduction of time needed for taxonomy gen- eration, as well as improvement of the taxonomy quality, compared to a manually created taxonomy. In Section 2 we describe some prior art and es- sential background. Section 3 describes our sug- gested taxonomy generation algorithm, while Sec- tion 4 describes the taxonomy editing tool. 2 Background 2.1 Taxonomy Generation Two major approaches to automatic domain taxon- omy generation can be identified in the literature. The first is the corpus-based approach, in which hierarchical clustering methods are applied either directly to keyword terms extracted from a target- domain corpus for generating a keyword hierarchy, or to the documents in the corpus with further ex- traction of category names as keywords frequent in …

Content truncated to 500 words