A Comparison of Graph Centrality Algorithms For Semantic Distance
Erhan Turan, Ersin Arslan, Çağatay Neftali Tülü, Umut Orhan · DergiPark (Istanbul University) · 2020
Semantic networks are kind of datasets used for natural language processing (NLP).Distance measurement for semantic networks, which are generally based on a graph structure, is a vital requirement for semantic analysis on concepts.Centrality measures can be used for calculating the semantic distance between concepts in a semantic network.In this paper, we evaluated graph centrality algorithms including PageRank, Hyperlink-induced Topic Search (HITS), and Betweenness Centrality on a semantic network, which was created from a Turkish dictionary dataset.Centrality measures special to these algorithms are used to calculate the semantic distance between synonym pairs in the semantic network.Also, we have used a simple centrality method beside the other three popular centrality algorithms to find out the most accurate and cost-effective method on our semantic network.Working on a bipartite model of the network which increases the complexity of implementation for centrality algorithms and performing calculations on a semantic network, that can be expanded with new nodes and edges, are two major challenges to overcome.Considering all these conditions, results from each algorithm are compared to pick out an optimal method for the semantic network.