Multihop Fuzzy Spatiotemporal RDF Knowledge Graph Query via Quaternion Embedding
Hao Ji, Li Yan, Zongmin Ma
Abstract
The proliferation of uncertain spatiotemporal data has led to an increasing demand for fuzzy spatiotemporal knowledge modeling in various applications. However, performing multi-hop query modeling on incomplete fuzzy spatiotemporal knowledge graphs (KGs) poses significant challenges. Recently, embedding-based multi-hop KG querying approaches have gained attention. Yet, these approaches often overlook KG uncertainty and spatiotemporal sensitivity, resulting in the neglect of fuzzy spatiotemporal information during multi-hop path reasoning. To address these challenges, we propose a embedding-based multi-hop query model for fuzzy spatiotemporal KG. We use quaternion to jointly embed spatiotemporal entities, and relations are represented as rotations from spatiotemporal subject to object. We incorporate uncertainty by the scoring function's bias factor, allowing for relaxation embedding. This approach facilitates the learning of a richer representation of fuzzy spatiotemporal KGs in vector space. By exploiting the inherent non-commutative compositional pattern of quaternions, we construct more accurate multi-hop paths within fuzzy spatiotemporal KGs, thus improving path reasoning performance. To evaluate the effectiveness of our model, we conduct experiments on two fuzzy spatiotemporal KG datasets, focusing on link prediction and path query answering. Results show that our proposed method significantly outperforms several state-of-the-art baselines in terms of performance metrics.