The article discusses the limitations of existing Temporal Knowledge Graph Question Answering (TKGQA) models. These models often fail to handle complex questions involving multiple temporal facts. To address this, the authors introduce the JointMultiFactsReasoning Network (JMFRN), a method that jointly reasons multiple temporal facts for accurately answering complex temporal questions. The JMFRN first retrieves question-related temporal facts from TKG for each entity of the given complex question. It then uses two different attention modules to aggregate entities and timestamps information of retrieved facts. The authors also introduce an additional answer type discrimination task to filter incorrect type answers. The proposed method shows significant improvement over existing models on the complex temporal question benchmark TimeQuestions.
Publication date: 5 Jan 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2401.02212
