Large-Scale Multi-granular Concept Extraction Based on Machine Reading Comprehension

The concepts in knowledge graphs (KGs) enable machines to understand natural language, and thus play an indispensable role many applications. However, existing KGs have the poor coverage of concepts, especially fine-grained concepts. In order supply with more new we propose a novel concept extraction framework, namely MRC-CE, extract large-scale multi-granular from descriptive texts entities. Specifically, MRC-CE is built machine reading comprehension model based on BERT, which can pointer network. Furthermore, random forest rule-based pruning are also adopted enhance MRC-CE's precision recall simultaneously. Our experiments evaluated upon multilingual KGs, i.e., English Probase Chinese CN-DBpedia, justify superiority over state-of-the-art models KG completion. Particularly, after running for each entity than 7,053,900 (instanceOf relations) supplied into KG. code datasets been released at https://github.com/fcihraeipnusnacwh/MRC-CE