Entity disambiguation based on parse tree neighbours on graph attention network

Entity disambiguation (ED) aims to link textual mentions in a document to the correct named entities in a knowledge base (KB). Although global ED model usually outperforms local model by collectively linking mentions based on the topical coherence assumption, it may still incur incorrect entity assignment when a document contains multiple topics. Therefore, we propose to extract global features locally, i.e., among a limited number of neighbouring mentions, to combine the respective superiority of both models. In particular, we derive mention neighbours according to the syntactic distance on a dependency parse tree, and propose a tree connection method CoSimTC to measure the cross-tree distance between mentions. Besides, we extend the Graph Attention Network (GAT) to integrate both local and global features to produce a discriminative representation for each candidate entity. Our experimental results on five widely-adopted public datasets demonstrate better performance compared with state-of-the-art approaches