The DOI matches its Crossref record; 49 of the 49 references it cites were found in a registry or an index; 7 of the 26 sources its claims cite were checked for support.
Toward Context-Aware Anomaly Detection for AIOps in Microservices Using Dynamic Knowledge Graphs
Open accessAbstract
This paper introduces context-aware anomaly detection using dynamic knowledge graphs for evolving monitored systems and presents an open-source benchmark with monitoring, fault injection, and data collection.
Overview
CoChat Summary
TL;DR
The paper proposes a context-aware anomaly detection methodology for microservices using dynamic knowledge graphs built from non-intrusive resource and network monitoring, plus an open benchmark with fault injection. It reports that context-aware models consistently outperform non-contextual baselines on a dynamic-topology benchmark.
Key findings
- The benchmark dataset contains roughly five days of monitoring data at 15-second granularity (24,770 time steps), with KGs averaging 1402 ± 109 triples per step, 39 injected faults, and 1,018 anomalous graphs.
- The best online context-aware model, INK-Z-ThresholdAD, achieves precision 0.786, recall 0.579, and F1 0.667, outperforming online baselines LSTM-AE (F1 0.186) and IF (F1 0.169).
- The offline INK-IF model reaches F1 0.466 versus IF F1 0.169, a 29.7% absolute F1 increase, and detects 89.5% of faults versus 26.3% for the baseline IF.
- A logical-OR ensemble increases recall up to 89.5% but lowers precision; the paper prioritizes high recall for early fault exposure in AIOps.
Methods
- The benchmark deploys Online Boutique on a Kubernetes cluster across seven Fed4Fire nodes, using Prometheus Node Exporter, cAdvisor, Istio, Locust for load generation, and Litmus Chaos for seven fault types (static and dynamic).
- Data are collected every 15 seconds and aggregated over five-minute windows; an OWL ontology models Kubernetes entities and monitoring literals, producing dynamic KGs with 250 unique entities, 847 unique relations, and 5,715,493 total triples.
- Anomaly detection uses INK neighborhood-based node embeddings, rolling-window Z-score aggregation, and detectors (QuantileAD, ThresholdAD, OutlierAD) with a logical-OR ensemble; evaluation uses a 20-40-40 split and compares against LSTM-AE and Isolation Forest baselines.
Limitations
- Stated: evaluation is conducted only on the Online Boutique application, which may limit generalizability of quantitative results; the paper frames this as a reproducible foundation for extension.
- Stated: a full evaluation of real-time deployment overhead is not included, though the pipeline is designed to be lightweight and fit within the 15-second monitoring interval.
- Stated: the best online model detects only 57.9% of faults and is tuned for higher precision; detected faults are mostly node-level, and the ensemble’s recall gain comes at a precision cost. Evident from the text: the dataset is collected in a research testbed with 39 injected faults, so class balance and production realism may differ from real deployments.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
5 Valid · 6 Not valid · 15 Not checked · 97 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“This design choice is consistent with operational expectations that in production microservice settings, anomalies typically need to be detected within 2-5 minutes to prevent cascading failures and minimize service impact [41].”
What the sources say
- Supportedcites [41] source
The case study (Section 5.3) states that anomalies in production microservice environments typically need to be detected within 2–5 minutes to prevent cascading failures and minimize service impacts. This directly supports the cited operational expectation. (LLM, conf=0.95)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“However, these scalability and resource efficiency benefits obtained through adoption of the MSA come at a cost of increased operational complexity in terms of monitoring and observability [4].”
What the sources say
- Supportedcites [4] source
The survey’s “Complexity of Traditional Monitoring Approaches” section says that microservices’ interconnected components make monitoring and root-cause analysis more difficult. Its discussion of monolithic architecture also says that debugging, performance management, and collecting observability data become more complex in a microservice architecture. Page numbers are not provided in the supplied passages. (LLM, conf=0.93)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“These solutions are based on fine-grained alerting and require multi-faceted thresholds for Key Performance Indicators (KPIs) in different layers of the application [8].”
What the sources say
- Partly supportedcites [8] sourceExperimental — accuracy not yet measured
Our copy does connect alerting with thresholds and tools with multiple application layers: §5.1 (p. 6) describes alerting as firing 'if some thresholds are triggered', and §7.3 Monitoring granularity (p. 12) documents tools operating at the application, microservice, VM/container and infrastructure levels. But the paper never characterises monitoring solutions as 'based on fine-grained alerting' nor says they 'require multi-faceted thresholds for KPIs' — it uses no 'KPI' construct at all (its taxonomy is user-oriented vs system-oriented metrics, §6.1–6.2, pp. 9–10), and it reports that less than 40% of the 71 surveyed tools have alerting features and that system-level metrics are much better supported than application/user-oriented ones, which runs against the claim's framing. The unsupported part is therefore the 'fine-grained alerting' basis and the requirement of multi-faceted KPI thresholds across layers. (LLM, conf=0.68)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The best performing model has an F1-score of 66.7%.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The detected faults, as discussed in Section VI, are limited to node-level faults.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
32 more claims
Create a free account to see all 123 claims and to verify the 2 references still open.
- Probably not a substantial claim. 56
- It may state two separate claims as one. 55
Citations
Cites
LOF: identifying density-based local outliers
Breunig, M. M., Kriegel, H.-P., Ng, R. T., & Sander, J. (2000) · ACM SIGMOD Record · doi:10.1145/335191.335388 · Cited by 5,207
LOF: identifying density-based local outliers
Breunig, M. M., Kriegel, H.-P., Ng, R. T., & Sander, J. (2000) · Proceedings of the 2000 ACM SIGMOD international conference on Management of data · doi:10.1145/342009.335388 · Cited by 4,149 · cited 1×
“While there are a multitude of univariate time-series AD models, three models are considered here, namely a percentile-based detector (QuantileAD), a threshold-based detector (ThresholdAD) and a local outlier detector (OutlierAD), which is a density-based technique for outlier identification [61].”
DeepLog: Anomaly Detection and Diagnosis from System Logs through Deep Learning
Du, M., Li, F., Zheng, G., & Srikumar, V. (2017) · Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security · doi:10.1145/3133956.3134015 · Cited by 1,835 · cited 1×
“DeepLog [16] uses a deep learning model, i.e., a Long Short-Term Memory (LSTM) model, to model a system log as a natural language sequence.”
Antoniou, G., & van Harmelen, F. (2004) · Handbook on Ontologies · doi:10.1007/978-3-540-24750-0_4 · Cited by 1,017
Drain: an online log parsing approach with fixed depth tree
He, P., Zhu, J., Zheng, Z., & Lyu, M. R. (2017) · 2017 IEEE International Conference on Web Services (ICWS) · doi:10.1109/icws.2017.13 · Cited by 943 · cited 1×
“These textual messages are often unstructured and require parsing strategies and systems [13].”
Microservices: The Journey So Far and Challenges Ahead
Jamshidi, P., Pahl, C., Mendonca, N. C., Lewis, J., & Tilkov, S. (2018) · IEEE Software · doi:10.1109/ms.2018.2141039 · Cited by 554 · cited 1×
“To ensure smooth operations and meet stakeholder demands, many companies have transitioned from a traditional monolithic architecture to a Microservice Architecture (MSA) [3].”
RDF2Vec: RDF Graph Embeddings for Data Mining
Ristoski, P., & Paulheim, H. (2016) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-46523-4_30 · Cited by 413 · cited 1×
“Traditional embedding techniques, such as TransE [57] and RDF2Vec [58], that reduce the dimensionality of the graph into fixed-length embeddings in the latent feature space cannot be applied efficiently to dynamic graphs for real-time or online prediction tasks.”
Representation learning for dynamic graphs: A survey
Kazemi, S. M., Goel, R., Jain, K., Kobyzev, I., Sethi, A., Forsyth, P., & Poupart, P. (2019) · arXiv · doi:10.48550/arxiv.1905.11485 · Cited by 338
Soldani, J., & Brogi, A. (2023) · ACM Computing Surveys · doi:10.1145/3501297 · Cited by 256 · cited 2×
“Related work in the domain of microservice-based monitoring and anomaly detection can be roughly divided based on the type of data used, i.e., log-based solutions, distributed tracing-based solutions and resource metric-based solutions [5].”
AIOps: Real-World Challenges and Research Innovations
Dang, Y., Lin, Q., & Huang, P. (2019) · 2019 IEEE/ACM 41st International Conference on Software Engineering: Companion Proceedings (ICSE-Companion) · doi:10.1109/icse-companion.2019.00023 · Cited by 238 · cited 1×
“Artificial Intelligence for IT Operations (AIOps) [9] aims to move from monitoring towards observability and address this issue through Machine Learning (ML).”
Research works printed but not linked: 21
- Brazil, 2018, Prometheus: Up & Running: Infrastructure and Application Performance Monitoring
- Lahmadi, Powering monitoring analytics with elk stack, 1
- 2024, Dynatrace: Unified observability and security
- 2024, Acme air sample and benchmark
- 2011, IEEE vast challenge
- 2024, Sock shop: A microservice demo application
- 2015, Online boutique: Cloud-first microservices demo application
- Calcote, 2019, Istio: Up and Running: Using a Service Mesh to Connect, Secure, Control, and Observe
- Sheikh, Modernize digital applications with microservices management using the istio service mesh, 359
- Heyman, 2024, Locust: Open-source performance and load testing
- Pierfederici, 2016, Distributed Computing with Python
- 2024, Litmuschaos: Open Source Chaos Engineering Platform
- 2021, AWS Fault Injection Service
- 2021, Azure Chaos Studio
- Butow, 2021, Chaos engineering: The history, principles, and practice
- 2024, Chaos mesh: Break your system constructively
- 2024, Argo: Workflow engine for kubernetes
- Burns, 2022, Kubernetes: Up and Running
- Soppelsa, 2016, Native Docker Clustering with Swarm
- Bordes, Translating embeddings for modeling multi-relational data, 1
- Analytics, 2019, A python toolkit for rule-based/unsupervised anomaly detection in time series
Cited by
Komarchesqui, M., Ruffo, V. G. D. S., Carvalho, L. F., & Proença, M. L. (2026) · Journal of Network and Systems Management · doi:10.1007/s10922-026-10084-z · Cited by 4
Jiang, Y., Liu, Y., Wang, X., Cheng, N., Jia, Z., Shi, D., Lv, Z., Li, Z., & Zhang, Y. (2026) · IEEE Transactions on Network and Service Management · doi:10.1109/tnsm.2026.3671463 · Cited by 1
“Among the enabling technologies, Knowledge Graph (KG) [3] is recognized as one of the effective means to achieve intelligent NOM, garnering significant attentions from researchers [4], [5], [6], [7], [8], [9].”
Tlili, S., Rahal, M., & Alghamdi, M. (2026) · World Wide Web · doi:10.1007/s11280-026-01425-8
Novi, D. D., Carnevale, L., & Villari, M. (2026) · IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid) · doi:10.1109/ccgrid68966.2026.00071
“Recent research has emphasized the importance of contextual awareness in dynamic microservice architectures, where topology, resource allocation, and service dependencies evolve continuously over time [13].”
NEKC: Few-Shot Anomaly Knowledge Completion for Intelligent Network Management
Zhang, Y., Rui, L., Wang, S., Zhang, J., Gao, Z., Qiu, X., & Guo, S. (2026) · IEEE Transactions on Network and Service Management · doi:10.1109/tnsm.2026.3724317
“To capture runtime dynamics, DKG-based context-aware anomaly detection models evolving dependencies among microservices, infrastructure resources, and monitoring observations, allowing anomaly detection to incorporate topology and context changes in dynamic microservice environments [27].”
Metadata
- Authors
- Moens, P., Steenwinckel, B., Ongenae, F., Volckaert, B., & Hoecke, S. V.
- Affiliations
- Ghent University (BE)
- Venue
- IEEE Transactions on Network and Service Management
- Published
- 2026
- Volume
- 23
- Pages
- 1970-1988
- ISSN
- 1932-4537 (ISSN-L)
- 2373-7379
- OpenAlex
- W7123344472
- Semantic Scholar
- db4735ef2276f7d689fe0ee911429e39dd4571ae
- Cited by
- 6
- References
- 41
- Citations by year
- 2026: 6
- FWCI
- 22.91
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Locations
- IEEE Transactions on Network and Service Management (publishedVersion, cc-by)
- Ghent University Academic Bibliography (Ghent University) (submittedVersion, cc-by)
- Funders
- Fonds Wetenschappelijk Onderzoek (1881020N)
- Topics
- Physical Sciences › Computer Science › Computer Networks and Communications › Software System Performance and Reliability
- Physical Sciences › Computer Science › Artificial Intelligence › Anomaly Detection Techniques and Applications
- Physical Sciences › Computer Science › Information Systems › Cloud Computing and Resource Management
- Keywords
- AIOps, microservices, dynamic knowledge graphs, observability, fault detection, infrastructure monitoring, benchmarking framework, fault injection
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Toward Context-Aware Anomaly Detection for AIOps in Microservices Using Dynamic Knowledge Graphs. https://app.cochat.ai/papers/10.1109/tnsm.2026.3652304
- The paper
- Moens, P., Steenwinckel, B., Ongenae, F., Volckaert, B., & Hoecke, S. V. (2026). Toward Context-Aware Anomaly Detection for AIOps in Microservices Using Dynamic Knowledge Graphs. IEEE Transactions on Network and Service Management. https://doi.org/10.1109/tnsm.2026.3652304
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.