MGCTQA: Multi-Granularity Complex Temporal Question Answering Based on Graph Neural Network and Knowledge Graph

To address the shortcomings of existing temporal question answering (QA) systems in handling complex semantic relations, cross-fact reasoning, and temporal information modeling, this paper proposes MGCTQA, a multi-granularity complex temporal QA model based on Graph Neural Networks (GNNs) and knowledge graphs. The model is designed with a hierarchical joint reasoning mechanism to comprehensively capture and deeply understand implicit entities and temporal constraints within questions. First, a pre-trained language model is used to obtain a global representation of the question. Next, a multi-hop entity-aware module applies GNNs to entity-related subgraphs to construct a complete reasoning chain. This is followed by a temporal information fusion module that thoroughly interprets the temporal constraints related to the question. Finally, a scoring function is employed to derive the final answer. Experimental results demonstrate that MGCTQA exhibits stable performance on simple questions and achieves significant performance gains on complex temporal queries.