Research on Predicting Super-Relational Data Links for Mine Hoists within Hyper-Relational Knowledge Graphs

Abstract

Hyper-relational knowledge graphs significantly enhance industrial production's intelligence, efficiency, and reliability by enabling equipment collaboration and optimizing supply chains. However, due to current limitations in data and technology, the construction of knowledge graphs in the industrial domain remains incomplete. Link prediction can effectively address this issue. This paper proposes a novel hyper-relational link prediction method called HyperFormer-LSTM, which integrates LSTM into the HyperFormer model and combines it with a MOE expert network to better capture information between entities. Experimental results show that this method performs excellently on both public datasets and self-constructed hoist datasets. In the MHSD (100) dataset, the MRR and H@1 increased by by 0.055 and 0.063, respectively, compared to HyperFormer. This method not only effectively solves the knowledge graph completion problem for mine hoists but also provides more accurate information for equipment maintenance and fault prediction. Key variables involved in this study include model structure, dataset type, dataset scale, evaluation metrics, and experimental settings. Future research will focus on further improving link prediction models and deeply studying the domain characteristics of mine hoist data to further advance research in the field of hyper-relational knowledge graph link prediction for mine hoist data.