Đang chuẩn bị nút TẢI XUỐNG, xin hãy chờ
Tải xuống
This paper describes an unsupervised knowledge–lean methodology for automatically determining the number of senses in which an ambiguous word is used in a large corpus. It is based on the use of global criterion functions that assess the quality of a clustering solution. | Selecting the Right Number of Senses Based on Clustering Criterion Functions Ted Pedersen and Anagha Kulkarni Department of Computer Science University of Minnesota Duluth Duluth MN 55812 USA tpederse kulka020 @d.umn.edu http senseclusters.sourceforge.net Abstract This paper describes an unsupervised knowledge-lean methodology for automatically determining the number of senses in which an ambiguous word is used in a large corpus. It is based on the use of global criterion functions that assess the quality of a clustering solution. 1 Introduction The goal of word sense discrimination is to cluster the occurrences of a word in context based on its underlying meaning. This is often approached as a problem in unsupervised learning where the only information available is a large corpus of text e.g. Pedersen and Bruce 1997 Schutze 1998 Pu-randare and Pedersen 2004 . These methods usually require that the number of clusters to be discovered k be specified ahead of time. However in most realistic settings the value of k is unknown to the user. Word sense discrimination seeks to cluster N contexts each of which contain a particular target word into k clusters where we would like the value of k to be automatically selected. Each context consists of approximately a paragraph of surrounding text where the word to be discriminated the target word is found approximately in the middle of the context. We present a methodology that automatically selects an appropriate value for k. Our strategy is to perform clustering for successive values of k and evaluate the resulting solutions with a criterion function. We select the value of k that is immediately prior to the point at which clustering does not improve significantly. Clustering methods are typically either parti-tional or agglomerative. The main difference is that agglomerative methods start with 1 or N clusters and then iteratively arrive at a pre-specified number k of clusters while partitional methods start by randomly .
