Lambda: Learning Matchable Prior For Entity Alignment with Unlabeled Dangling Cases
This addresses a practical challenge in knowledge graph integration for data management and AI applications, offering a solution for handling unlabeled dangling entities, though it is incremental in improving existing methods.
The paper tackles the entity alignment problem with unlabeled dangling cases, where some entities lack counterparts in other knowledge graphs, by proposing the Lambda framework with KEESA for alignment and iPULE for detection, achieving superior performance even when baselines use extra labeled data.
We investigate the entity alignment (EA) problem with unlabeled dangling cases, meaning that partial entities have no counterparts in the other knowledge graph (KG), and this type of entity remains unlabeled. To address this challenge, we propose the framework \textit{Lambda} for dangling detection and then entity alignment. Lambda features a GNN-based encoder called KEESA with spectral contrastive learning for EA and a positive-unlabeled learning algorithm for dangling detection called iPULE. iPULE offers theoretical guarantees of unbiasedness, uniform deviation bounds, and convergence. Experimental results demonstrate that each component contributes to overall performances that are superior to baselines, even when baselines additionally exploit 30\% of dangling entities labeled for training.