An Efficient Embedding Framework for Uncertain Attribute Graph

Graph data with uncertain connections between entities is commonly represented using uncertain graphs. This paper tackles the challenge of graph embedding within such uncertain attribute graphs. Current graph embedding techniques are typically oriented towards deterministic graphs, or uncertain graphs that lack attribute data. Furthermore, the majority of studies on uncertain graph learning simply adapt conventional algorithms for deterministic graphs to handle uncertainty, leading to compromised computational efficiency. To address these issues, we introduce an optimized embedding framework UAGE for uncertain attribute graphs. In UAGE, nodes are represented within a Gaussian distribution space to learn node attributes. We also propose a Probability Similarity Value (PSV) to manage relationship uncertainty and ensure that nodes with higher-order similar structures are located more closely in the latent space. Real-world dataset experiments confirm that UAGE surpasses contemporary methods in performance for downstream tasks.