The article presents a new approach to answering complex temporal questions over a Temporal Knowledge Graph (TKG) using a JointMultiFactsReasoning Network (JMFRN). Traditional TKG question answering models struggle with questions containing multiple temporal facts. JMFRN addresses this by retrieving question-related temporal facts for each entity in a complex question and using two different attention modules to aggregate entities and timestamps information. The method outperforms state-of-the-art solutions on the TimeQuestions benchmark.
Publication date: 5 Jan 2024
Project Page: unknown
Paper: https://arxiv.org/pdf/2401.02212
