Fuzzy K-means Clustering on a High Dimensional Semantic Space

摘要

One way of representing semantics is via a high dimensional conceptual space constructed from lexical co-occurrence. Concepts (words) are represented as a vector whereby the dimensions are other words. As the words are represented as dimensional objects, clustering techniques can be applied to compute word clusters. Conventional clustering algorithms, e.g., the K-means method, however, normally produce crisp clusters, i.e., an object is assigned to only one cluster. This is sometimes not desirable. Therefore, a fuzzy membership function can be applied to the K-Means clustering, which models the degree of an object belonging to certain cluster. This paper introduces a fuzzy k-means clustering algorithm and how it is used to word clustering on the high dimensional semantic space constructed by a cognitively motivated semantic space model, namely Hyperspace Analogue to Language. A case study demonstrates the method is promising.