An entity linking method using multiple Doc2Vec

Abstract

Appropriately linking a polysemous word in a text to its corresponding entity in a knowledge base is an essential part in producing structured knowledge. Linking becomes challenging as issues such as name variations, entity ambiguities or absence of entity in knowledge base occurs in the process. We present an entity linking method consisting of multiple Doc2Vec model, achieving high performance and cost effectiveness. In the experiments, our proposed method achieved 83.5% in mapping accuracy, improving 31.0 points higher than the simple string matching.