Concept Induction via Fuzzy C-Means Clustering in a High Dimensional Semantic Space

Lexical semantic space models have recently been investigated to automatically derive the meaning (semantics) of information based on natural language usage. In a semantic space, a term can be considered as a concept represented geometrically as a vector, the components of which correspond to terms in a vocabulary. A primary way to perform reasoning in a semantic space is to categorize concepts in the space into a number of regions (i.e., groups). Such a process is referred to as concept induction, which can be realized by clustering objects in the space. The resulting groups can potentially form a basis for knowledge discovery and ontology construction. Conventional clustering algorithms, e.g., the K-Means method, normally produce crisp clusters, i.e., an object could be assigned to only one cluster. It is not always the case in reality. For example, a word “Reagan” may belong to both the cluster about administration of US government, and another one about the Iran-contra scandal. Therefore, a membership function is applied, which determines the degree to which an object belongs to different clusters. This chapter introduces a cognitively motivated semantic space model, namely Hyperspace Analogue to Language (HAL), and shows how a fuzzy C-Means clustering algorithm is used to concept categorization in the high dimensional semantic space. The experimental results indicate that applying fuzzy C-Means clustering over the HAL semantic space is promising in constructing semantically related groups of terms.