Reasoning over Temporal Knowledge Graphs (TKGs) aims to forecast future events based on historical ones. Existing approaches typically enforce strict temporal order constraints among past events; however, such rigidity limits the effective exploitation of path information during reasoning, thereby reducing both model flexibility and predictive performance. To address this limitation, this paper introduces SR-RTR (Sub-graph Reasoning based on Relaxed Temporal Relation), an interpretable subgraph reasoning framework designed to fully harness path information within TKGs. By incorporating a relaxed temporal factor, the proposed method softens the chronological constraints on historical events, broadens the sampling scope of candidate nodes during subgraph reasoning, and enhances the efficiency of path information utilization. This mechanism reflects human cognitive intuition: when the temporal gap between two events falls within a certain threshold, their sequence can be considered interchangeable. SR-RTR constructs a query-specific inference subgraph from the TKG and iteratively performs two core operations—subgraph expansion and pruning—until the entity with the highest attention score is identified as the prediction result. Extensive experiments on four benchmark datasets demonstrate that SR-RTR uncovers a greater number of reasoning paths relevant to the target prediction, leading to substantial improvements in both reasoning accuracy and computational efficiency.

Claim extraction and decontextualization are fundamental steps in automated fact-checking tasks. Claim extraction aims to identify core check-worthy claims from unverified text, while decontextualization seeks to eliminate the contextual dependencies of core claims so that they can be understood independently of the original text. In practical applications, information to be verified is usually presented in the form of documents, and existing document-level claim extraction and decontextualization methods often rely on complex multi-model architectures. This paper proposes LLM-CED, a document-level claim extraction and decontextualization method based on large language models, which consists of four modules: core-claim extraction, ambiguous-unit identification and question generation, question-answering-driven decontextualization, and self-reflective review. By using a single general-purpose LLM as the shared underlying model across all stages, LLM-CED reduces model heterogeneity and simplifies the integration and coordination of different processing stages. Experiments on the AVeriTeC-DCE dataset show that LLM-CED outperforms the evaluated baseline methods in document-level claim extraction, decontextualization, and the overall task. For the overall document-level claim extraction and decontextualization task, LLM-CED achieves a chrF score of 33.52% under the top-three evaluation setting.