Context-aware entity harmonization mechanisms
OntoCast solves the problem of entity designation ambiguity by means of BERT-based vector space model, and its technical implementation contains three key points: 1) constructing a contextual feature library to record the semantic environment in which the entity appears; 2) realizing a disambiguation algorithm based on the attention mechanism, which is able to differentiate between the category of technology companies or fruits when detecting "Apple"; 3) establishing a cross-document entity mapping, such as automatically associating different authors' expressions of "deep learning" in a collection of academic papers. When "Apple" is detected, it can distinguish between technology companies or fruit categories; 3) build cross-document entity mapping, such as automatically correlating different authors' expressions of "deep learning" in a collection of academic papers. The test data shows that in the medical literature processing scenario, the technology improves the entity recognition accuracy from 82% to 96%.
This answer comes from the articleOntoCast: an intelligent framework for extracting semantic triples from documentsThe
