Multihop Knowledge Chain Query on Fine-Grained Fuzzy Spatiotemporal Knowledge Graph by Embedding

Performing multi-hop queries on a large-scale Knowledge Graph (KG) presents significant computational challenges. Although subgraph matching-based methods show ability, they return empty/few answers when querying missing KG. An emerging solution involves embedding both entities and multi-hop queries of KG into a vector space, where the proximity between queries and entities reflects their semantic relevance. However, current embedding-based methods fail to adequately address three critical aspects: temporal semantics, spatial semantics, and fine-grained fuzzy semantics in multi relational queries. Thus, we propose a multi-hop query method on fuzzy spatiotemporal KG (STKG) by fine-grained embedding. We first develop an element-level embedding mechanism that maps fuzzy spatiotemporal semantics into each component. We then construct ordered multi-hop relations as composable operations on paths by Hermitian dot products. We finally use a Path-Constraint Resource Allocation (PCRA) to model path reliability. In addition, we validate our method on fuzzy spatiotemporal datasets. Results show that the method outperforms several state-of-the-art models and can effectively handle fuzzy spatiotemporal knowledge chain queries.