The paper "Joint Multi-Facts Reasoning Network for Complex Temporal Question Answering Over Knowledge Graph" addresses the challenge of answering complex temporal questions in Temporal Knowledge Graphs (TKGs). Existing models struggle with questions containing multiple temporal facts, often relying on the assumption that each question contains only a single temporal fact. To tackle this, the authors propose the Joint Multi-Facts Reasoning Network (JMFRN), which retrieves and reasons over multiple temporal facts to provide accurate answers. Key contributions of JMFRN include: 1. **Fact Retrieval**: JMFRN retrieves relevant temporal facts from the TKG for each entity in the question. 2. **Attention Modules**: Two attention modules (Entity-aware and Time-aware) are designed to aggregate entity and timestamp information from retrieved facts. 3. **Answer Type Discrimination**: An additional task is introduced to filter out incorrect answer types, enhancing model performance and training stability. Experiments on the TimeQuestions dataset demonstrate that JMFRN significantly outperforms state-of-the-art models, particularly in handling complex questions with multiple entities. The method's effectiveness is validated through ablation studies and performance comparisons with various baselines.The paper "Joint Multi-Facts Reasoning Network for Complex Temporal Question Answering Over Knowledge Graph" addresses the challenge of answering complex temporal questions in Temporal Knowledge Graphs (TKGs). Existing models struggle with questions containing multiple temporal facts, often relying on the assumption that each question contains only a single temporal fact. To tackle this, the authors propose the Joint Multi-Facts Reasoning Network (JMFRN), which retrieves and reasons over multiple temporal facts to provide accurate answers. Key contributions of JMFRN include: 1. **Fact Retrieval**: JMFRN retrieves relevant temporal facts from the TKG for each entity in the question. 2. **Attention Modules**: Two attention modules (Entity-aware and Time-aware) are designed to aggregate entity and timestamp information from retrieved facts. 3. **Answer Type Discrimination**: An additional task is introduced to filter out incorrect answer types, enhancing model performance and training stability. Experiments on the TimeQuestions dataset demonstrate that JMFRN significantly outperforms state-of-the-art models, particularly in handling complex questions with multiple entities. The method's effectiveness is validated through ablation studies and performance comparisons with various baselines.