TN-AutoRCA: Benchmark Construction and Agentic Framework for Self-Improving Alarm-Based Root Cause Analysis in Telecommunication Networks
Keyu Wu, Qianjin Yu, Manlin Mei, Ruiting Liu, Jun Wang, Kailai Zhang, Yelun Bao
TL;DR
This paper introduces TN-AutoRCA, a benchmark and framework for self-improving alarm-based root cause analysis in telecom networks, addressing AI challenges in complex graph reasoning and benchmark scarcity.
Contribution
It presents a novel benchmark and agentic framework for self-improving RCA, enhancing AI capabilities in complex telecommunication network analysis.
Findings
Developed a new benchmark for RCA in telecom networks
Proposed an agentic framework for self-improvement in RCA
Demonstrated improved accuracy in root cause identification
Abstract
Root Cause Analysis (RCA) in telecommunication networks is a critical task, yet it presents a formidable challenge for Artificial Intelligence (AI) due to its complex, graph-based reasoning requirements and the scarcity of realistic benchmarks.
Peer Reviews
Decision·Submitted to ICLR 2026
1. New Benchmark: TN-RCA is the benchmark for RCA in telecom networks, addressing a gap in the field. It is built on real-world data and features a principled difficulty stratification method via cycle consistency checks. 2. Agentic Framework Design: Auto-RCA introduces a modular, feedback-driven agent architecture that systematically improves code-based solutions.
1. Limited Motivational Depth: While RCA is an important industrial problem, the paper does not sufficiently justify why it is a meaningful or challenging AI research task. The connection to broader machine learning challenges—beyond telecom-specific applications—is underdeveloped, making the contribution feel more applied than foundational. 2. Superficial Experimental Analysis: The experiments, though comprehensive in model comparisons, lack in-depth analysis. For instance, there is little di
1. The benchmark and experiments are grounded in real-world telecommunication scenarios and data. 2. The paper demonstrates that a code-based solution can be effective for RCA tasks, achieving strong results on OpenRCA.
1. Novelty and relevance: The Auto-RCA framework appears applicable to general code-based problem-solving tasks; it is unclear which design elements are specifically tailored to the telecommunication RCA context. 2. Missing statistics: The paper lacks details about the training set’s size and difficulty distribution, which are crucial given that Auto-RCA’s core mechanism relies on contrastive feedback. 3. Missing case analysis: The paper is missing failure cases of LLM-only methods and pre-/post
S1: The proposed benchmark is the first RCA benchmark in telecommunication networks.
W1: The paper lacks focus. It proposed a new benchmark TN-RCA, which should be challenging. And then it proposed a new method Auto-RCA, which achieves a 91% F1-score. It seems that the new benchmark is not challenging enough. W2: For the proposed benchmark TN-RCA, the authors only evaluated LLM-based methods, but ignored the traditional RCA methods. RCA for graph-structured data has been extensively studied in the microservice domain [1][2][3]. It is not clear how they perform on these benchmark
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsSoftware System Performance and Reliability · Network Security and Intrusion Detection · Smart Grid Security and Resilience