Enhancing statistical semantic networks with concept hierarchies

With the emergence of the semantic web, effective knowledge representation has gained importance. Statistically generated semantic networks are simple representations whose semantic power is yet to be completely explored. Though, these semantic networks are created with simple statistical measures without much overhead, they have the potential to express the semantic relationship between concepts. In this paper, we explore the capability of such networks and enhance them with concept hierarchies to serve as better knowledge representations. The concept hierarchies are built based on the level of importance of concepts. The level of importance/coverage of a concept within the given set of documents has to be taken into account to build an effective knowledge representation. In this paper, we provide a domain-independent, graph based approach for identifying the level of importance of each concept from the statistically generated semantic network which represents the entire document set. Insights about the depth of every concept is obtained by analysing the graph theoretical properties of the statistically generated semantic network. A generic concept hierarchy is created using a greedy strategy, and the original semantic network is reinforced with this concept hierarchy. Experiments over different data sets demonstrate that our approach works effectively in classifying concepts and generating taxonomies based on it, thereby effectively enhancing the semantic network.