Multihop Fuzzy Event Knowledge Graph Reasoning With Event Soft Logical Rules Guidance

An event refers to an occurrence under a specific topic, at a particular time and location. With the growing presence of uncertain event information, there is increasing demand for modeling fuzzy event knowledge in real-world applications. However, performing multi-hop reasoning over fuzzy event knowledge graphs (EKGs) remains challenging due to the uncertainty and dynamic nature of events. Recent multi-hop reasoning methods based on knowledge graphs (KGs) have shown promise but often neglect event uncertainty and dynamics, limiting their effectiveness in fuzzy event reasoning. To address this, we propose a novel model, MFEGR, for multi-hop reasoning over fuzzy event knowledge graphs (FEKGs), guided by soft logical rules. MFEGR explores logical connections among multi-hop paths in EKGs and discovers underlying soft rules governing them. Rule-based regularization is introduced to learn both temporal-causal dependencies between events and semantic relationships between events and their arguments. Additionally, we enhance event embeddings by incorporating fine-grained fuzziness into event elements and coarse-grained fuzziness into soft logic rules, improving reasoning robustness. We evaluate MFEGR on three fuzzy EKG datasets for event link prediction and multi-hop reasoning. Experimental results show that MFEGR consistently outperforms some state-of-the-art baselines across multiple evaluation metrics.