Researchers have developed a new framework called SABET-QA for answering questions based on temporal knowledge graphs. These graphs contain time-sensitive facts and relationships between entities over time. Existing methods struggle to reason about these complex queries in multiple steps. SABET-QA addresses this issue by iteratively refining its reasoning states across multiple hops using a bidirectional scoring mechanism and contextualization module. The framework also incorporates a differentiable working memory for progressive hypothesis refinement and auxiliary temporal boundaries for coarse supervision when available. Experiments on several datasets show consistent improvements over strong baselines, particularly on complex multi-step queries. --- Why it matters: This matters to AI researchers because it tackles the challenge of reasoning about time-sensitive facts in knowledge graphs, which is crucial for applications like event forecasting, decision-making under uncertainty, and natural language processing. Source: https://arxiv.org/abs/2608.20083