Right Answer at the Right Time - Temporal Retrieval-Augmented Generation via Graph Summarization

TL;DR

STAR-RAG tackles time-consistent QA on temporal knowledge graphs by introducing a training-free retrieval framework that builds a time-aligned rule graph and employs seeded personalized PageRank to focus evidence on temporally relevant neighborhoods. The method uses an MDL-based edge selection to compress the graph and reduce search space, and seeds propagation with a blend of corpus coverage and query-based ranking to prioritize temporally aligned evidence. Across CronQuestion, Forecast, and MultiTQ, STAR-RAG achieves higher answer accuracy while drastically lowering token usage compared to strong GraphRAG baselines, demonstrating effective temporal reasoning with improved efficiency. This approach enables scalable, deployment-friendly temporal RAG without fine-tuning, offering practical impact for real-world temporal QA tasks and heavy reasoning over evolving knowledge graphs.

Abstract

Question answering in temporal knowledge graphs requires retrieval that is both time-consistent and efficient. Existing RAG methods are largely semantic and typically neglect explicit temporal constraints, which leads to time-inconsistent answers and inflated token usage. We propose STAR-RAG, a temporal GraphRAG framework that relies on two key ideas: building a time-aligned rule graph and conducting propagation on this graph to narrow the search space and prioritize semantically relevant, time-consistent evidence. This design enforces temporal proximity during retrieval, reduces the candidate set of retrieval results, and lowers token consumption without sacrificing accuracy. Compared with existing temporal RAG approaches, STAR-RAG eliminates the need for heavy model training and fine-tuning, thereby reducing computational cost and significantly simplifying deployment.Extensive experiments on real-world temporal KG datasets show that our method achieves improved answer accuracy while consuming fewer tokens than strong GraphRAG baselines.

Figures (6)

• Figure 1: Comparison of vanilla GraphRAG and STAR-RAG. Vanilla GraphRAG relies primarily on semantic matching for retrieval and ignores explicit temporal constraints, which degrades answer accuracy. STAR-RAG maps events to a time-aligned rule graph and performs propagation on this graph to narrow the search space and prioritize time-consistent evidence, enabling temporally aligned retrieval and improved performance.
• Figure 2: Running example of building the rule graph based on MultiTQ.
• Figure 3: Comparison of token consumption and reasoning time based on MultiTQ.
• Figure 4: The average accuracy (%, denoted as Acc.) and reasoning time (s, denoted as Tim.) for each variant of STAR-RAG.
• Figure 5: Performance of setting different $K_1$ and $K_2$ based on MultiTQ.