PolarisX(2): auto-growing context-aware knowledge graph

Abstract

Artificial intelligence requires advanced technologies in various fields. In particular, natural language processing consists of various tasks because computers need to understand and process human languages. Knowledge graphs represent common sense as a graph, making it easy to understand the relationships between entities. Various studies exist because knowledge graphs could play a crucial role in computers' understanding of natural language. PolarisX is an auto-growing knowledge graph that could especially cope with neologisms. However, existing studies have a limitation in that they rarely correspond to information containing numbers representing a cardinal, ordinal, or quantity and can extract only one relationship from one sentence. We propose the auto-growing context-aware knowledge graph, PolarisX(2), an entity extraction model capable of responding to numeric information, and a relation extraction model considering type. It also enables multiple knowledge extraction from a single sentence by applying the candidate pair construction model.