Adaptive Retrieval Agents for Factual and Cost-Aware Cyber Threat Intelligence Summarization

View at publisher Summarise with VIRAT
Authors: Mona Rajhans, Vishal Khawarey
Cited by: 0
References: 2

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Cyber threat intelligence (CTI) text is large, heterogeneous, and difficult for analysts to summarize accurately. While Retrieval-Augmented Generation (RAG) improves factual grounding in large language models (LLMs), current systems typically rely on a fixed retrieval depth, causing unnecessary latency or inconsistent factuality. This paper addresses this challenge by introducing an adaptive retrieval agent that adjusts context size in real time using a retrieval-aware factuality metric, the Grounded Entailed Claim Score (GECS). Retrieval selection is cast as a cost-utility optimization problem,$J=G-\lambda \tau$, enabling the agent to balance factual gain ($G$) against computational cost ($\tau$) without modifying or retraining the underlying LLM. Experiments on two cybersecurity corpora-structured NVD vulnerabilities and semi-structured MITRE ATT&CK narratives- show that the agent matches optimal fixed-depth performance on structured text while avoiding costly retrieval when factual benefits are limited. These results demonstrate that factuality metrics can serve as lightweight internal feedback signals for self-calibrating and cost-efficient RAG systems in real-world CTI workflows.

Fields of Study & Subjects

Topic Modeling Information Retrieval and Search Behavior Adversarial Robustness in Machine Learning

Publication details

Journal / source
International Conference on Advanced Communications Technology (ICACT)
Pages
577–582
Publication date
8 February 2026
Language
EN