Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 45 of the 45 references it cites were found in a registry or an index; 5 of the 42 sources its claims cite were checked for support.

BRITD: behavior rhythm insider threat detection with time awareness and user adaptation

Open access
Song, Shuang, Gao, Neng, Zhang, Yifei, Ma, Cunqing · 2024 · Cybersecurity

Abstract

This paper proposes a behavioral feature extraction scheme with time awareness and user adaptation for insider threat detection using deep learning.

Overview

CoChat Summary

TL;DR

BRITD is a deep-learning insider threat detection scheme that implicitly encodes absolute time in behavior feature sequences and adapts sequence construction per user via covariance. On CMU CERT v6.2 it reports overall AUC 0.9730 and precision 0.8072, outperforming all baselines.

Key findings

  • BRITD achieves overall AUC 0.9730 and precision 0.8072 on CMU CERT v6.2; the paper states this is at least 0.0802 higher in AUC and at least 0.2746 higher in precision than baselines.
  • The user-adaptive feature construction selects the `<G, b>` scheme maximizing feature-sequence covariance; selected optimal schemes were Scen.1 `<12 h, 7 h>`, Scen.2 `<12 h, 8 h>`, Scen.3 `<12 h, 5 h>`, Scen.4 `<12 h, 8 h>`, Scen.5 `<12 h, 8 h>`.
  • Ablation shows stacked BiLSTM helps contextual and collective threat scenarios, while FNN improves detection of point, coarse-grained collective, and fine-grained collective threat behaviors; BRITD is best in most scenarios, with Scenario 4 showing slight overfitting because its threats are time-independent point threats.
  • Runtime is reported as under 3 s total test time for all test instances of a single user; training time ranges from 426.71 s at 1-h granularity to 65.86 s at 24-h granularity, with 78.49 s at the selected 12-h granularity.

Methods

  • Feature extraction keeps user id and time, converts other log attributes to behavior tokens, aggregates logs per user chronologically, divides 24 h into periods of granularity `G` with starting bias `b`, counts tokens per period into a 129-dimension statistical vector, and concatenates daily vectors chronologically so absolute time becomes position information.
  • The detection model is an autoencoder: encoder uses stacked BiLSTM plus FNN, decoder uses FNN; loss is mean squared reconstruction error, training is semi-supervised on data without insider threat instances, and reconstruction error is the outlier score.
  • Experiments use CMU CERT v6.2 (4,000 employees over 516 days); a subset where each user generated insider threat behaviors covering all five scenarios is used, with the first 150 days per user for training and the rest for testing; metrics are AUC and precision (max precision when recall = 1), thresholded by the minimum anomaly value of true anomalies, evaluated at 24-h/day granularity and averaged over 10 replicates.

Limitations

  • Paper-stated: detection granularity is not fine enough; accuracy and precision, especially in Scenario 2, still have room for improvement; user behavioral rhythms may change over time but the adaptive scheme does not account for this; and the CERT dataset is synthetic, making real-world effectiveness difficult to demonstrate.
  • Paper-stated: optimal threshold selection still depends on the proportion of insider threat instances because the number of test instances is not large enough; adding self-attention can overfit when time granularity is larger and sequences are shorter; Scenario 4 shows slight overfitting because its threats are time-independent.
  • Evident from Table 1: some scenario test sets are small (e.g., Scenario 3 has 7 test days), which may limit robustness of the per-scenario results.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1186/s42400-023-00190-9

Cites

45 of 45 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

5 Valid · 5 Not valid · 29 Not checked · 95 Own results and interpretations

70 more claims

Create a free account to see all 134 claims and to verify the 15 references still open.

See all claims — free account
  • Probably not a substantial claim. 34
  • It may state two separate claims as one. 34
  • The quote only partly states this claim. 2

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 43 references

Research works printed but not linked: 2

  1. Alsaheel A, Nan Y, Ma S, Yu L, Walkup G, Celik ZB, Zhang X, Xu D (2021) ATLAS: A sequence-based learning approach for attack investigation. In: USENIX security symposium
  2. Schlkopf B, Williamson RC, Smola A, Shawe-Taylor J, Platt J (1999) Support vector method for novelty detection. Adv Neural Inf Process Syst, 12

Cited by

See all 45

Metadata

Authors
Song, S., Gao, N., Zhang, Y., & Ma, C.
Affiliations
Chinese Academy of Sciences (CN)
Corresponding author
Gao, Neng
Venue
Cybersecurity
Published
2024-01-02
Volume
7
Issue
1
ISSN
2523-3246 (ISSN-L)
OpenAlex
W4390486486
Cited by
45
References
43
Citations by year
2026: 24 · 2025: 14 · 2024: 7
FWCI
16.95
Type
journal article
Language
en
Open access
Licence
Locations
  • Cybersecurity (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
National Key Research and Development Program of China
Topics
  • Physical Sciences › Computer Science › Information Systems › Information and Cyber Security
  • Physical Sciences › Computer Science › Computer Networks and Communications › Network Security and Intrusion Detection
  • Physical Sciences › Computer Science › Information Systems › User Authentication and Security Systems
Keywords
insider threat detection, user behavior analysis, user adaptation, stacked bidirectional LSTM

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).

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: BRITD: behavior rhythm insider threat detection with time awareness and user adaptation. https://app.cochat.ai/papers/10.1186/s42400-023-00190-9
The paper
Song, S., Gao, N., Zhang, Y., & Ma, C. (2024). BRITD: behavior rhythm insider threat detection with time awareness and user adaptation. Cybersecurity. https://doi.org/10.1186/s42400-023-00190-9

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.