Domain-specific knowledge graph is constructed based on domain knowledge. The knowledge graph consists of entities and relations, where relation construction is the core. Traditional domain-specific knowledge graph relation construction requires manual construction by experts familiar with the relevant domain, which is heavily labor cost consuming. To solve this problem, this paper proposes a knowledge graph relation construction model ASRC based on active learning and semantic recognition, which consists of relation existence mining module and relation category prediction module. ASRC is training to mine entity pairs with possible relation, and recommends the entity pairs to experts for relation labeling, and then predicts the specific category of relation. ASRC improves model quality through multiple iterations, and more relation annotation can be labeled with the same number of annotations compared to randomly selected entity pairs.