AISLES THROUGH THE CATEGORY FOREST: Utilising the Wikipedia Category System for Corpus Building in Machine Learning

摘要

The Word Wide Web is a continuous challenge to machine learning. Established approaches have to be enhanced and new methods be developed in order to tackle the problem of finding and organising relevant information. It has often been motivated that semantic classifications of input documents help solving this task. But while approaches of supervised text categorisation perform quite well on genres found in written text, newly evolved genres on the web are much more demanding. In order to successfully develop approaches to web mining, respective corpora are needed. However, the composition of genre- or domain-specific web corpora is still an unsolved problem. It is time consuming to build large corpora of good quality because web pages typically lack reliable meta information. Wikipedia along with similar approaches of collaborative text production offers a way out of this dilemma. We examine how social tagging, as supported by the MediaWiki software, can be utilised as a source of corpus building. Further, we describe a representation format for social ontologies and present the Wikipedia Category Explorer, a tool which supports categorical views to browse through the Wikipedia and to construct domain specific corpora for machine learning.