Right Answer at the Right Time - Temporal Retrieval-Augmented Generation via Graph Summarization (2510.16715v1)

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.

• The paper presents STAR-RAG, a framework that integrates temporal graph summarization to enhance temporal question answering.
• It utilizes a personalized PageRank mechanism over a time-aligned rule graph, ensuring retrieval of time-consistent and accurate evidence.
• Experimental results show a 97% reduction in token usage and a 9.1% improvement in accuracy compared to conventional methods.

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

Introduction

The paper "Right Answer at the Right Time - Temporal Retrieval-Augmented Generation via Graph Summarization" presents STAR-RAG, a novel retrieval-augmented generation framework specifically designed to handle temporal question answering over knowledge graphs. Unlike conventional RAG systems, which often neglect temporal constraints, STAR-RAG integrates a temporal graph summarization strategy that enhances both efficiency and accuracy in retrieving time-consistent answers.

Temporal Graph Summarization

The fundamental contribution of STAR-RAG lies in its ability to construct a time-aligned rule graph. This approach involves summarizing temporal knowledge into a structured format that facilitates efficient propagation of queries. The critical aspect of this method is the construction of candidate rule nodes and the establishment of edges representing temporal relationships between events.

Figure 1: Running example of building the rule graph based on MultiTQ.

Propagation and Evidence Retrieval

STAR-RAG employs a propagation mechanism using personalized PageRank on the rule graph to effectively narrow the search space to time-consistent evidence. This process ensures that the retrieved information aligns temporally with the query constraints, thereby improving the factual accuracy of generated answers.

Figure 2: Comparison of vanilla GraphRAG and STAR-RAG. Vanilla GraphRAG relies on semantic matching, while STAR-RAG uses a time-aligned rule graph for improved performance.

Performance Evaluation

Extensive experiments demonstrate the superior performance of STAR-RAG compared to existing temporal GraphRAG approaches. The framework achieves a remarkable improvement in answer accuracy, with a significant reduction in token usage. For instance, compared to the MedicalGraphRAG, STAR-RAG reduces token consumption by 97% while improving accuracy by 9.1%.

Figure 3: Comparison of token consumption and reasoning time based on MultiTQ.

Moreover, the studies highlight the robustness of STAR-RAG across various datasets involving complex temporal reasoning tasks.

Ablation Studies

The paper also provides insights from ablation studies to underscore the importance of the rule graph and temporal alignment strategies. Variants of STAR-RAG without rule graph utilization show a notable drop in performance, emphasizing the value of this structural approach to understanding temporal dynamics.

Conclusion

STAR-RAG represents an advancement in retrieval-augmented generation methodologies by addressing temporal constraints effectively. Its deployment across different LLM backbones indicates its adaptability and efficiency without the need for extensive retraining. This work opens pathways for future research into more sophisticated temporal reasoning systems and sets a benchmark for efficient token utilization and accurate answer retrieval in temporal knowledge graphs.