Entity linking plays a vital role in Question Answering over Knowledge Graphs (KGQA), and the representation of entities is a fundamental component of entity linking for user questions. In order to alleviate the problem of entity descriptions that unrelated texts obfuscate similar entities, we present a new entity linking framework, which refines the encodings of entity descriptions based on entity difference modeling, so that entity linking’s ability to distinguish among similar entities is improved. The entity differences are modeled in a two-stage approach: the initial differences are first computed among similar entity candidates by comparing their descriptions, and then interactions between the initial differences and questions are performed to extract key differences, which identify critical information for entity linking. On the basis of the key differences, subsequent entity description encodings are refined, and entity linking is then performed using the refined entity representations. Experimental results on end-to-end benchmark datasets demonstrate that our approach achieves state-of-the-art precision, recall and F1-score.
CITATION STYLE
Wang, M., Li, M., Sun, K., & Hou, Z. (2022). Entity Difference Modeling Based Entity Linking for Question Answering over Knowledge Graphs. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 13551 LNAI, pp. 221–233). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-17120-8_18
Mendeley helps you to discover research relevant for your work.