Author(s): José Eduardo Talavera Herrera, Javier Guillot Jimenez, Marco Antonio Casanova, Bernardo Pereira Nunes, Giseli Rabello Lopes, Luiz André P. Paes Leme
Abstract: This paper presents an empirical evaluation of a family of path search strategies that help understand the connectivity of a pair of entities in a knowledge graph. Each path search strategy in the family combines an entity similarity measure and a path ranking measure to find and order relevant paths between a given pair of entities. The paper adopts a benchmark from two entertainment domains, and a baseline from the literature. It applies the pairwise comparison method to compare the path search strategies with the benchmark, and with the baseline. Finally, the paper shows that the path search strategy that adopts the Jaccard index and the Exclusivity-based Relatedness ranking measure was the best strategy, and obtained better results than the baseline.
Keywords: Entity relatedness; Entity similarity measure; Path ranking measure