Overall Model Architecture Overview

论文上下文

Paper title: AdapTime: Enabling Adaptive Temporal Reasoning in Large Language Models Abstract: Large language models have demonstrated strong reasoning capabilities in general knowledge question answering. However, their ability to handle temporal information remains limited. To address this limitation, existing approaches often involve external tools or manual verification and are tailored to specific scenarios, leading to poor generalizability. Moreover, these methods apply a fixed pipeline to all questions, overlooking the fact that different types of temporal questions require distinct reasoning strategies, which leads to unnecessary processing for simple cases and inadequate reasoning for complex ones. To this end, we propose AdapTime, an adaptive temporal reasoning method that dynamically executes reasoning steps based on the input context. Specifically, it involves three temporal reasoning actions: reformulate, rewrite and review, with an LLM planner guiding the reasoning process. AdapTime integrates seamlessly with state-of-the-art LLMs and significantly enhances their temporal reasoning capabilities without relying on external support. Extensive experiments demonstrate the effectiveness of our approach. Passages referencing this figure: 2023 ); Kamalloo et al. ( 2023 ); Dong et al. ( 2024a ); Jia et al. ( 2025 ); Zhang et al. ( 2026 ) . However, they still face significant challenges in handling temporal questions Son and Oh ( 2023 ); Xiong et al. ( 2024 ); Deng et al. ( 2025 ) . Temporal reasoning tasks are concerned with understanding documents containing temporal information and answering time-sensitive questions. As shown in Figure 1 , answering a question such as “ Which position did Terence Cooper hold between Mar 1966 and Oct 1969? ” requires both retrieving relevant facts and aligning them with the temporal context. In such cases, LLMs often generate incorrect answers due to a poor understanding of temporal expressions (e.g., “between Mar 1966 and Oct 1969”) or difficulty in identifying the order of events Chu et ? ” requires both retrieving relevant facts and aligning them with the temporal context. In such cases, LLMs often generate incorrect answers due to a poor understanding of temporal expressions (e.g., “between Mar 1966 and Oct 1969”) or difficulty in identifying the order of events Chu et al. ( 2023 ) . Therefore, enhancing the temporal reasoning capabilities of LLMs remains a critical challenge. Figure 1: An example of temporal reasoning in question answering. Recent approaches typically focus on developing reasoning pipelines involve analyzing temporal questions, extracting timelines, and verifying generated answers Xiong et al. ( 2024 ); Bazaga et al. ( 2025 ) . While these reasoning strategies enhance the model’s ability to handle temporal information, they often rely on external suppo mple of temporal reasoning in question answering. Recent approaches typically focus on developing reasoning pipelines involve analyzing temporal questions, extracting timelines, and verifying generat