Detecting Sexual Predatory Chats by Perturbed Data and Balanced Ensembles
Open accessAbstract
Securing the safety of the children on online platforms is critical to avoid the mishaps of them being abused for sexual favors, which usually happens through predatory conversations. A number of approaches have been proposed to analyze the content of the messages to identify predatory conversations. However, due to the non-availability of large-scale predatory data, the state-of-the-art works employ a standard dataset that has less than 10% predatory conversations. Dealing with such heavy class imbalance is a challenge to devise reliable predatory detection approaches. We present a new approach for dealing with class imbalance using a hybrid sampling and class re-distribution to obtain an augmented dataset. To further improve the diversity of classifiers and features in the ensembles, we also propose to perturb the data along with augmentation in an iterative manner. Through a set of experiments, we demonstrate an improvement of 3% over the best state-of-the-art approach and results in an F1-score of 0.99 and an Fβof 0.94 from the proposed approach.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Some papers attempted to consider this setting and have tried to address this as an imbalanced data problem.
“Some papers attempted to consider this setting and have tried to address this as an imbalanced data problem [62,72,73,86].”
From Online grooming detection: A comprehensive survey of child exploitation in chat logs (2023) · cites this paper as [86]
The data for grooming detection is highly imbalanced, negatively impacting the accuracy metric's relevance.
“For instance, the data for grooming detection is highly imbalanced, negatively impacting the accuracy metric's relevance [86].”
From Online grooming detection: A comprehensive survey of child exploitation in chat logs (2023) · cites this paper as [86]
The imbalanced nature of grooming data will lead to a sub-optimal classifier that gives more weight to one class over the other and results in underfitting or oversampling.
“The imbalanced nature of grooming data will lead to a sub-optimal classifier that gives more weight to one class over the other and results in underfitting or oversampling [86].”
From Online grooming detection: A comprehensive survey of child exploitation in chat logs (2023) · cites this paper as [86]
While the F1-score of 0.96 is relatively lower than Borj et al. [8], the F0.5-score is higher, making the approach more applicable in law-enforcement cases.
“While the 𝐹 1 -score of 0.96 is relatively lower than Borj et al. [8], we note that 𝐹 0.5 -score is higher, making the approach more applicable in law-enforcement cases.”
From Detecting Online Grooming By Simple Contrastive Chat Embeddings (2023) · cites this paper as [8]
Borj et al. [8] tried to handle the imbalanced nature of grooming data by hybrid sampling and class re-distribution and gained an accuracy of 0.99 and an F0.5-score of 0.94.
“For instance, Borj et al. [8] tried to handle the imbalanced nature of grooming data by hybrid sampling and class re-distribution and gained an accuracy of 0.99 and an 𝐹 0.5 -score of 0.94.”
From Detecting Online Grooming By Simple Contrastive Chat Embeddings (2023) · cites this paper as [8]
The citation lists haven’t been fetched yet
References 25 · Cited by 13
- Authors
- Borj, P. R., Raja, K., & Bours, P.
- Affiliations
- Norwegian University of Science and Technology (NO)
- Venue
- 2021 International Conference of the Biometrics Special Interest Group (BIOSIG)
- Publisher
- IEEE
- Published
- 2021-09
- Pages
- 1-5
- OpenAlex
- W3202576386
- Cited by
- 13
- References
- 25
- Citations by year
- 2026: 3 · 2025: 2 · 2024: 1 · 2023: 4 · 2022: 3
- FWCI
- 0.92
- Type
- conference paper
- Language
- en
- Open access
- green
- hdl.handle.net
- Locations
- International Conference of the Biometrics Special Interest Group (BIOSIG) (publishedVersion)
- NORA - Norwegian Open Research Archives (submittedVersion, other-oa)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Hate Speech and Cyberbullying Detection
- Physical Sciences › Computer Science › Artificial Intelligence › Imbalanced Data Classification Techniques
- Physical Sciences › Computer Science › Information Systems › Spam and Phishing Detection
- Keywords
- online chat, child online safety, class imbalance, data augmentation, perturbed data, hybrid sampling, message content analysis
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.
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
