Set-based visualization and enhancement of embedding results for heterogeneous multi-label networks

Heterogeneous networks are ubiquitous in the real-world, such as social networks and brain cell networks. Network embedding techniques have emerged as powerful tools for generating low-dimensional representations of nodes in information networks, and have become an effective method for analyzing and mining heterogeneous networks. However, these embeddings can still be too high-dimensional for human perception. To address this issue, dimensionality reduction techniques are often used to project the embeddings onto a two-dimensional plane, enabling visual analysis through scatterplots. However, previous research on the interpretation and enhancement of dimensionality-reduced data has rarely focused on more complex multi-labeled data. In this paper, focusing on the high-dimensional multi-label embedding results of heterogeneous networks, we propose a set-based visualization and enhancement approach, which constructs the set space for multi-label attributes, and links the embedding space to the set space, to solve the multi-label complexity problem and provide enhancement techniques that specialize in different tasks. The effectiveness and applicability of our proposed approach are extensively validated through expert interview, user experiment and two case studies on real-world datasets.