A comprehensive evaluation of entropy-based directionality estimation method

In network systems, directional information is crucial, but constructing directional networks presents significant challenges. Recently, a method for extracting directionality from undirected networks based on information theory has been proposed. However, a comprehensive assessment of the efficacy of this approach has been delayed. In this study, we utilized the World Trade Web (WTW) network, employed in the original paper, to assess the method’s validity with various Rényi entropy parameters and three different edge removals based on ascending, descending, and random sorting of edge weights. Our results demonstrate that the initial evaluation on WTW resulting in high accuracy and precision is correlated with a substantial number of reciprocal edges, hence the high-level performance scores decrease when edge removal occurs. In contrast, the entropy-based directionality extraction method achieved low precision for the entire brain networks of Drosophila larva, and this network contains a limited number of reciprocal edges. These findings call for a caution and systematic performance evaluation when employing the entropy-based directionality extraction method.