No abstract

Fig. 1. BIASCOPE is an interactive tool for visual unfairness diagnosis for graph embeddings. The tool consists of three views: (A) a Statistical Summary View, (B) an Unfairness Comparison View and (C) a Diagnose View. The content of the views is as follows: (A) The Statistical Summary View summarizes key properties of the selected network. (B) The Unfairness Comparison View is a side-by-side view that facilitates the fairness comparison of two embeddings of the same network. Nodes are colored based on their unfairness score, where darker nodes are embedded more unfairly. For example, the figure shows that the observed bias can be evenly distributed throughout the network or concentrated in certain communities. (C) The Diagnose View shows the embedding subspace affecting the unfairness score of a selected node, together with its ego network. The two visualizations are interactively linked, allowing the user to investigate how bias appears in an embedding and how it relates to the relevant graph structure.