Are Negative Links Really Beneficial to Network Embedding?: In-Depth Analysis and Interesting Results

In this paper, we start by pointing out the limitations on the validation of existing signed network embedding (NE) methods. To address the limitations, we design the two research questions: (1) are signed NE methods consistently more effective in various types of tasks than unsigned NE methods? (2) in signed NE methods, does the utilization of negative links help provide higher accuracy in various tasks? To answer the questions, we present our evaluation framework consisting of three components: (1) five signed network datasets; (2) six signed and two unsigned NE methods; (3) five types of tasks. Through extensive experiments on our evaluation framework, we demonstrate that additional utilization of negative links really helps only in some tasks related to negative links but not in tasks related to positive links.