Modeling a Hierarchical Abstraction Process on top of Co-Occurrence Graphs

Abstract

A co-occurrence graph is incorporated from sets of documents that represent knowledge. However, determining number of groups or clusters of knowledge this may pertain to remains a challenge. This work will explore the hierarchical clustering algorithm for which a hierarchy is built from the cluster center (centroid) of each cluster that is read node by node. Each node finds an inter-cluster that will be assigned by referring to a distance from the node to the inter-cluster center which ensures that this node is a member of that inter-cluster. The inter-cluster center is an abstract identifier that represents all nodes of the respective cluster. When the next hierarchy level is built; the clustering will be applied again. All processes are repeated until the last remaining abstract identifier (root). The results of 10 datasets showed that the co-occurrence graph can be hierarchical clustering for which the hierarchical levels ended at level 4.