Optimizing laboratory safety education knowledge graph through neural-symbolic reasoning

Aiming at the limitation that the laboratory safety education knowledge graph (KG) lacks the ability to mine symbolic rules and is difficult to reason about the implicit relationships in the KG, this paper first represents the entities and relations in the laboratory safety education KG, and then designs neural network encoding and symbolic rule encoding to extract neural-symbolic features, respectively. Contrastive learning technology is used to project the neural network knowledge embedding and symbolic rule knowledge embedding of the same sample into the same vector space, and to bring the same samples closer and push the different samples further apart in the same vector space, achieving further fusion of neural-symbolic features. Experimental results show that the MRR of the proposed model is 0.956, which is better than the comparative model, proving that the model has excellent performance.