Fuzzy Spatiotemporal Knowledge Graph Queries via Global and Local Uncertain Dynamic Subgraph Reasoning Using Embedding

A large amount of work focus on static and temporal knowledge graph (KG) queries via subgraph reasoning, but it fails to consider the space-sensitivity and fuzziness of KG. Subgraph query in fuzzy spatiotemporal knowledge graph (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">f</i>-STKG) is a fundamental and challenging task. Although a few efforts exist dedicated to <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">f</i>-STKG queries based on subgraph matching, they are sensitive to missing information in the knowledge base. To this end, we propose the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">f</i>-STKG query method by subgraph reasoning using embeddings. The method aims to capture the semantic similarity between the query subgraph and candidate subgraph of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">f</i>-STKG to accomplish subgraph matching. We first embed <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">f</i>-STKGs into vector space to form <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">f</i>-STKG vectors. Then, we construct a global subgraph of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">f</i>-STKG using Composition-based Graph Convolutional Neural Networks (CompGCN), which allows global information adjacent to this subgraph to be integrated. Subsequently, the query subgraph of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">f</i>-STKG is constructed using mean pooling and Gated Recurrent Unit (GRU), as well as the candidate subgraph of the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">f</i>-STKG is constructed using CompGCN and GRU. This facilitates the integration of structural and sequence pattern information of subgraphs. Finally, we measured the semantic similarity between this query subgraph and the candidate subgraph using the subgraph matching function. In addition, we designed experiments to validate the effectiveness of our method on two datasets. Results show that our method outperforms state-of-the-art baselines.