pith. sign in

arxiv: 2505.20654 · v2 · submitted 2025-05-27 · 💻 cs.CL · cs.AI

Chinese Cyberbullying Detection: Dataset, Method, and Validation

Yi Zhu , Xin Zou , Xindong Wu This is my paper

Pith reviewed 2026-05-19 13:14 UTC · model grok-4.3

classification 💻 cs.CL cs.AI
keywords cyberbullying detectionChinese datasetincident detectionpseudo labelingensemble methodsocial media analysisannotation validationincident prediction
0
0 comments X

The pith

A new annotation method builds the first Chinese cyberbullying dataset organized by incidents rather than comment polarity.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Existing cyberbullying benchmarks focus on the polarity of individual comments such as offensive or non-offensive, which essentially performs hate speech detection. This paper instead organizes data around real-world incidents that draw widespread social attention. It uses an ensemble of three explanation-based detection methods to generate pseudo labels, which human annotators then review using proposed evaluation criteria. The resulting CHNCI dataset contains 220,676 comments across 91 incidents and serves as a benchmark for both cyberbullying detection and incident prediction tasks.

Core claim

The paper establishes that combining three cyberbullying detection methods based on explanations generation into an ensemble produces pseudo labels of usable quality, which human annotators can judge to construct CHNCI, the first Chinese cyberbullying incident detection dataset consisting of 220,676 comments in 91 incidents, and that this dataset functions as a benchmark for the tasks of cyberbullying detection and incident prediction.

What carries the argument

Ensemble of three explanation-generation cyberbullying detectors that produce pseudo labels for human validation into incident-level data with new evaluation criteria.

If this is right

  • Supplies a benchmark dataset specifically for incident-level cyberbullying detection in Chinese.
  • Supports research on predicting whether a stream of comments will become a cyberbullying incident.
  • Introduces explicit criteria for deciding when a collection of comments constitutes a cyberbullying incident.
  • Moves analysis away from isolated comment polarity toward incident organization for more realistic modeling.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The pseudo-labeling ensemble approach could be reused to build incident-organized datasets for other languages or online harms.
  • Incident-level data may integrate more naturally with social-media monitoring systems that track unfolding events.
  • Future experiments could test whether adding user metadata or image features to this dataset further improves prediction performance.

Load-bearing premise

The pseudo labels from the ensemble of three methods are accurate enough that human annotators can reliably turn them into a valid incident-level dataset.

What would settle it

Independent human review finding that a substantial fraction of the pseudo labels are incorrect, or models trained on CHNCI showing no improvement in incident prediction accuracy over models trained on polarity-based datasets.

Figures

Figures reproduced from arXiv: 2505.20654 by Xindong Wu, Xin Zou, Yi Zhu.

Figure 1
Figure 1. Figure 1: The overview of our method for building Chinese cyberbullying detection dataset organized by incidents. The data are collected [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Category distribution of the CHNCI dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Performance Comparison with Baseline Methods [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The process of dataset validation. Step 1: Scraping comments related to cyberbullying incidents, which may include offensive [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Hourly Trend of Comments: Comparison of Cyberbullying Incidents and Normal Events. The x-axis represents the hours elapsed [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Word clouds of online comments during the event. (a) shows cyberbullying content, and (b) shows non-cyberbullying content. The [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
read the original abstract

Existing cyberbullying detection benchmarks were organized by the polarity of speech, such as "offensive" and "non-offensive", which were essentially hate speech detection. However, in the real world, cyberbullying often attracted widespread social attention through incidents. To address this problem, we propose a novel annotation method to construct a cyberbullying dataset that organized by incidents. The constructed CHNCI is the first Chinese cyberbullying incident detection dataset, which consists of 220,676 comments in 91 incidents. Specifically, we first combine three cyberbullying detection methods based on explanations generation as an ensemble method to generate the pseudo labels, and then let human annotators judge these labels. Then we propose the evaluation criteria for validating whether it constitutes a cyberbullying incident. Experimental results demonstrate that the constructed dataset can be a benchmark for the tasks of cyberbullying detection and incident prediction. To the best of our knowledge, this is the first study for the Chinese cyberbullying incident detection task.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript introduces CHNCI, the first Chinese cyberbullying incident detection dataset consisting of 220,676 comments across 91 incidents. It describes a novel annotation pipeline that ensembles three explanation-generation-based cyberbullying detection methods to produce pseudo-labels, which human annotators then judge, along with proposed evaluation criteria for validating cyberbullying incidents. The authors claim the resulting dataset serves as a benchmark for cyberbullying detection and incident prediction tasks.

Significance. If the pseudo-label quality and human annotation reliability can be demonstrated, the work would offer a meaningful contribution by shifting from polarity-based hate-speech detection to incident-organized data, which better captures real-world cyberbullying dynamics. As the first such Chinese resource, it could support new research on incident prediction; the explanation-based ensemble approach is a potentially useful innovation for scalable labeling.

major comments (3)
  1. [Abstract and §3] Abstract and §3 (Method): The ensemble of three explanation-generation cyberbullying detectors is described at a high level, but no precision, recall, F1, or other quantitative metrics are reported for the pseudo-labels generated before human review. This directly undermines verification that the pseudo-labels are of sufficient quality to support reliable incident-level annotation of 220k comments.
  2. [§4] §4 (Annotation and Validation): No inter-annotator agreement statistics (e.g., Cohen’s or Fleiss’ kappa) or comparison between pseudo-labels and final human labels are provided. Given that the central claim rests on human judgment producing a trustworthy benchmark dataset, these metrics are load-bearing for assessing label noise and dataset validity.
  3. [§5] §5 (Experiments): The claim that the dataset constitutes a benchmark for detection and prediction tasks lacks reported baseline comparisons, specific performance numbers on the proposed evaluation criteria, or ablation on the ensemble’s contribution. Without these, the experimental validation of the dataset’s utility remains unsubstantiated.
minor comments (2)
  1. [Abstract] The abstract would be clearer if it briefly named the three detection methods and the ensemble aggregation rule rather than referring to them generically.
  2. Consider adding a table summarizing dataset statistics (e.g., comments per incident, label distribution) to improve readability and allow quick assessment of scale and balance.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below, indicating where revisions will be made to improve the manuscript's rigor and transparency.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (Method): The ensemble of three explanation-generation cyberbullying detectors is described at a high level, but no precision, recall, F1, or other quantitative metrics are reported for the pseudo-labels generated before human review. This directly undermines verification that the pseudo-labels are of sufficient quality to support reliable incident-level annotation of 220k comments.

    Authors: We agree that quantitative metrics for the pseudo-label generation step are necessary to allow readers to assess quality prior to human review. The current manuscript describes the ensemble at a high level without reporting precision, recall, or F1. In the revised manuscript, we will add these metrics in §3, computed on a held-out validation set for both individual methods and the ensemble, to substantiate the pseudo-label quality. revision: yes

  2. Referee: [§4] §4 (Annotation and Validation): No inter-annotator agreement statistics (e.g., Cohen’s or Fleiss’ kappa) or comparison between pseudo-labels and final human labels are provided. Given that the central claim rests on human judgment producing a trustworthy benchmark dataset, these metrics are load-bearing for assessing label noise and dataset validity.

    Authors: We acknowledge that inter-annotator agreement statistics and pseudo-to-final label comparisons are critical for demonstrating annotation reliability and quantifying label noise. These are not reported in the current version. We will incorporate Cohen’s and Fleiss’ kappa values, along with agreement analysis between pseudo-labels and human labels, into the revised §4 to strengthen validation of the dataset. revision: yes

  3. Referee: [§5] §5 (Experiments): The claim that the dataset constitutes a benchmark for detection and prediction tasks lacks reported baseline comparisons, specific performance numbers on the proposed evaluation criteria, or ablation on the ensemble’s contribution. Without these, the experimental validation of the dataset’s utility remains unsubstantiated.

    Authors: We recognize that additional experimental details are needed to fully substantiate the benchmark utility. While the manuscript includes experimental results on detection and prediction, it lacks explicit baselines, specific numbers on the criteria, and ensemble ablations. In the revised §5, we will add standard baseline comparisons, report concrete performance figures, and include an ablation on the ensemble to provide stronger validation. revision: yes

Circularity Check

0 steps flagged

No significant circularity in dataset construction or benchmark claims

full rationale

The paper constructs the CHNCI dataset from external comments by first generating pseudo labels via an ensemble of three explanation-based cyberbullying detectors, then applying human annotator judgment and newly proposed incident-level evaluation criteria. This process relies on independent external data sources and human review rather than any self-referential definitions, fitted parameters renamed as predictions, or load-bearing self-citations. No equations or uniqueness theorems reduce the central claims to tautological inputs, and the experimental demonstrations on the resulting dataset do not loop back to the construction steps by construction. The derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the unstated effectiveness of the explanation-based ensemble for pseudo-labeling and on the human annotators' ability to produce reliable incident labels using the proposed criteria; these are domain assumptions rather than derived results.

axioms (1)
  • domain assumption Cyberbullying incidents can be identified and validated through a combination of automated explanation-generating detectors and subsequent human judgment.
    This premise underpins the entire dataset construction process described in the abstract.

pith-pipeline@v0.9.0 · 5692 in / 1305 out tokens · 63198 ms · 2026-05-19T13:14:55.619453+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages · 3 internal anchors

  1. [1]

    Cyberbullying classification meth- ods for arabic: A systematic review

    [ALBayari et al., 2021] Reem ALBayari, Sharif Abdullah, and Said A Salloum. Cyberbullying classification meth- ods for arabic: A systematic review. In The International Conference on Artificial Intelligence and Computer Vision, pages 375–385. Springer,

    work page 2021

  2. [2]

    Image cyberbullying de- tection and recognition using transfer deep machine learn- ing

    [Almomani et al., 2024] Ammar Almomani, Khalid Nahar, Mohammad Alauthman, Mohammed Azmi Al-Betar, Qus- sai Yaseen, and Brij B Gupta. Image cyberbullying de- tection and recognition using transfer deep machine learn- ing. International Journal of Cognitive Computing in En- gineering, 5:14–26,

    work page 2024

  3. [3]

    An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling

    [Bai et al., 2018] Shaojie Bai, J Zico Kolter, and Vladlen Koltun. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271,

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  4. [4]

    Cyberbullying detection and machine learning: a systematic literature review

    [Balakrisnan and Kaity, 2023] Vimala Balakrisnan and Mo- hammed Kaity. Cyberbullying detection and machine learning: a systematic literature review. Artificial Intel- ligence Review, 56(Suppl 1):1375–1416,

    work page 2023

  5. [5]

    Topicality of cyberbully- ing among teenagers in russia and latvia

    [Boronenko et al., 2013] Vera Boronenko, Vladimir Men- shikov, and Gilberto Marzano. Topicality of cyberbully- ing among teenagers in russia and latvia. Social Sciences Bulletin, 1(16):84–104,

    work page 2013

  6. [6]

    Cyberbullying detection: Utilizing so- cial media features

    [Bozyi˘git et al., 2021] Alican Bozyi ˘git, Semih Utku, and Efendi Nasibov. Cyberbullying detection: Utilizing so- cial media features. Expert Systems with Applications , 179:115001,

    work page 2021

  7. [7]

    Hatebert: Retrain- ing bert for abusive language detection in english

    [Caselli et al., 2020] Tommaso Caselli, Valerio Basile, Je- lena Mitrovi´c, and Michael Granitzer. Hatebert: Retrain- ing bert for abusive language detection in english. arXiv preprint arXiv:2010.12472,

    work page arXiv 2020

  8. [8]

    Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

    [Cho, 2014] Kyunghyun Cho. Learning phrase representa- tions using rnn encoder-decoder for statistical machine translation. arXiv preprint arXiv:1406.1078,

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  9. [9]

    Comparing zero-shot text classification and rule-based matching in identifying cyberbullying behav- iors on social media

    [Chong et al., 2022] Wei Jiek Chong, Hui Na Chua, and May Fen Gan. Comparing zero-shot text classification and rule-based matching in identifying cyberbullying behav- iors on social media. In 2022 IEEE International Confer- ence on Artificial Intelligence in Engineering and Technol- ogy (IICAIET), pages 1–5. IEEE,

    work page 2022

  10. [10]

    A coefficient of agreement for nominal scales

    [Cohen, 1960] Jacob Cohen. A coefficient of agreement for nominal scales. Educational and psychological measure- ment, 20(1):37–46,

    work page 1960

  11. [11]

    Cold: A benchmark for chinese offensive language detec- tion

    [Deng et al., 2022] Jiawen Deng, Jingyan Zhou, Hao Sun, Chujie Zheng, Fei Mi, Helen Meng, and Minlie Huang. Cold: A benchmark for chinese offensive language detec- tion. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing , pages 11580– 11599. Association for Computational Linguistics,

    work page 2022

  12. [12]

    Is this a violation? learning and understanding norm violations in online com- munities

    [dos Santos et al., 2024] Thiago Freitas dos Santos, Nardine Osman, and Marco Schorlemmer. Is this a violation? learning and understanding norm violations in online com- munities. Artificial Intelligence, 327:104058,

    work page 2024

  13. [13]

    The Llama 3 Herd of Models

    [Dubey et al., 2024] Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783,

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  14. [14]

    Traditional bullying and cyberbullying among chil- dren and adolescents in germany–cross-sectional results of the 2017/18 hbsc study and trends

    [Fischer et al., 2020] Saskia M Fischer, Nancy John, Wolf- gang Melzer, Anne Kaman, Kristina Winter, and Ludwig Bilz. Traditional bullying and cyberbullying among chil- dren and adolescents in germany–cross-sectional results of the 2017/18 hbsc study and trends. Journal of health mon- itoring, 5(3):53,

    work page 2020

  15. [15]

    Measuring nominal scale agreement among many raters

    [Fleiss, 1971] Joseph L Fleiss. Measuring nominal scale agreement among many raters. Psychological bulletin , 76(5):378,

    work page 1971

  16. [16]

    Long short-term memory

    [Hochreiter, 1997] S Hochreiter. Long short-term memory. Neural Computation MIT-Press,

    work page 1997

  17. [17]

    Knowledgeable prompt-tuning: Incorpo- rating knowledge into prompt verbalizer for text classifi- cation

    [Hu et al., 2021] Shengding Hu, Ning Ding, Huadong Wang, Zhiyuan Liu, Jingang Wang, Juanzi Li, Wei Wu, and Maosong Sun. Knowledgeable prompt-tuning: Incorpo- rating knowledge into prompt verbalizer for text classifi- cation. arXiv preprint arXiv:2108.02035,

    work page arXiv 2021

  18. [18]

    Chain of explanation: New prompting method to gen- erate quality natural language explanation for implicit hate speech

    [Huang et al., 2023] Fan Huang, Haewoon Kwak, and Jisun An. Chain of explanation: New prompting method to gen- erate quality natural language explanation for implicit hate speech. In Companion Proceedings of the ACM Web Con- ference 2023, pages 90–93,

    work page 2023

  19. [19]

    Cyberbullying detection solutions based on deep learn- ing architectures

    [Iwendi et al., 2023] Celestine Iwendi, Gautam Srivastava, Suleman Khan, and Praveen Kumar Reddy Maddikunta. Cyberbullying detection solutions based on deep learn- ing architectures. Multimedia Systems, 29(3):1839–1852,

    work page 2023

  20. [20]

    Language model-based approach for multiclass cyberbullying detection

    [Kaddoura and Nassar, 2025] Sanaa Kaddoura and Reem Nassar. Language model-based approach for multiclass cyberbullying detection. In International Conference on Web Information Systems Engineering , pages 78–89. Springer,

    work page 2025

  21. [21]

    Bert: Pre-training of deep bidirectional transformers for lan- guage understanding

    [Kenton and Toutanova, 2019] Jacob Devlin Ming- Wei Chang Kenton and Lee Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for lan- guage understanding. In Proceedings of naacL-HLT , volume 1, page

    work page 2019

  22. [22]

    A human-centered systematic literature review of cy- berbullying detection algorithms

    [Kim et al., 2021] Seunghyun Kim, Afsaneh Razi, Gianluca Stringhini, Pamela J Wisniewski, and Munmun De Choud- hury. A human-centered systematic literature review of cy- berbullying detection algorithms. Proceedings of the ACM on Human-Computer Interaction, 5(CSCW2):1–34,

    work page 2021

  23. [23]

    Conprompt: Pre- training a language model with machine-generated data for implicit hate speech detection

    [Kim et al., 2023] Youngwook Kim, Shinwoo Park, Young- soo Namgoong, and Yo-Sub Han. Conprompt: Pre- training a language model with machine-generated data for implicit hate speech detection. In Findings of the Associ- ation for Computational Linguistics: EMNLP 2023, pages 10964–10980,

    work page 2023

  24. [24]

    Cyberbullying and cyber-mobbing in developing countries

    [Kintonova et al., 2021] Aliya Kintonova, Alexander Vasyaev, and Viktor Shestak. Cyberbullying and cyber-mobbing in developing countries. Information & Computer Security, 29(3):435–456,

    work page 2021

  25. [25]

    A bi-gru with attention and capsnet hybrid model for cyberbullying detection on social media

    [Kumar and Sachdeva, 2022] Akshi Kumar and Nitin Sachdeva. A bi-gru with attention and capsnet hybrid model for cyberbullying detection on social media. World Wide Web, 25(4):1537–1550,

    work page 2022

  26. [26]

    Gpt understands, too

    [Liu et al., 2024] Xiao Liu, Yanan Zheng, Zhengxiao Du, Ming Ding, Yujie Qian, Zhilin Yang, and Jie Tang. Gpt understands, too. AI Open, 5:208–215,

    work page 2024

  27. [27]

    Detection of harassment type of cyberbul- lying: A dictionary of approach words and its impact

    [Mahbub et al., 2021] Syed Mahbub, Eric Pardede, and ASM Kayes. Detection of harassment type of cyberbul- lying: A dictionary of approach words and its impact. Security and Communication Networks, 2021(1):5594175,

    work page 2021

  28. [28]

    Cyberbullying detec- tion for low-resource languages and dialects: Review of the state of the art

    [Mahmud et al., 2023] Tanjim Mahmud, Michal Ptaszynski, Juuso Eronen, and Fumito Masui. Cyberbullying detec- tion for low-resource languages and dialects: Review of the state of the art. Information Processing & Manage- ment, 60(5):103454,

    work page 2023

  29. [29]

    Mtbullygnn: a graph neural network-based multitask framework for cyberbully- ing detection

    [Maity et al., 2022] Krishanu Maity, Tanmay Sen, Sriparna Saha, and Pushpak Bhattacharyya. Mtbullygnn: a graph neural network-based multitask framework for cyberbully- ing detection. IEEE Transactions on Computational Social Systems, 11(1):849–858,

    work page 2022

  30. [30]

    A machine learn- ing approach to cyberbullying detection in arabic tweets

    [Musleh et al., 2024] Dhiaa Musleh, Atta Rahman, Mo- hammed Abbas Alkherallah, Menhal Kamel Al-Bohassan, Mustafa Mohammed Alawami, Hayder Ali Alsebaa, Jawad Ali Alnemer, Ghazi Fayez Al-Mutairi, May Issa Aldossary, Dalal A Aldowaihi, et al. A machine learn- ing approach to cyberbullying detection in arabic tweets. Computers, Materials & Continua, 80(1),

    work page 2024

  31. [31]

    Kpt++: Refined knowledgeable prompt tuning for few-shot text classification

    [Ni and Kao, 2023] Shiwen Ni and Hung-Yu Kao. Kpt++: Refined knowledgeable prompt tuning for few-shot text classification. Knowledge-Based Systems , 274:110647,

    work page 2023

  32. [32]

    Natural language watermarking via paraphraser-based lexical substitution

    [Qiang et al., 2023] Jipeng Qiang, Shiyu Zhu, Yun Li, Yi Zhu, Yunhao Yuan, and Xindong Wu. Natural language watermarking via paraphraser-based lexical substitution. Artificial Intelligence, 317:103859,

    work page 2023

  33. [33]

    Cy- berbullying detection with weakly supervised machine learning

    [Raisi and Huang, 2017] Elaheh Raisi and Bert Huang. Cy- berbullying detection with weakly supervised machine learning. In Proceedings of the 2017 IEEE/ACM Interna- tional Conference on Advances in Social Networks Analy- sis and Mining 2017, pages 409–416,

    work page 2017

  34. [34]

    Cyberbul- lying detection: Hybrid models based on machine learning and natural language processing techniques

    [Raj et al., 2021] Chahat Raj, Ayush Agarwal, Gnana Bharathy, Bhuva Narayan, and Mukesh Prasad. Cyberbul- lying detection: Hybrid models based on machine learning and natural language processing techniques. Electronics, 10(22):2810,

    work page 2021

  35. [35]

    Using machine learning to detect cyberbullying

    [Reynolds et al., 2011] Kelly Reynolds, April Kontostathis, and Lynne Edwards. Using machine learning to detect cyberbullying. In 2011 10th International Conference on Machine learning and applications and workshops , vol- ume 2, pages 241–244. IEEE,

    work page 2011

  36. [36]

    A “deeper” look at detecting cyberbullying in social networks

    [Rosa et al., 2018] Hugo Rosa, David Matos, Ricardo Ribeiro, Luisa Coheur, and Jo ˜ao P Carvalho. A “deeper” look at detecting cyberbullying in social networks. In 2018 international joint conference on neural networks (IJCNN), pages 1–8. IEEE,

    work page 2018

  37. [37]

    Automatic cyberbullying detection: A systematic review

    [Rosa et al., 2019] Hugo Rosa, N ´adia Pereira, Ricardo Ribeiro, Paula Costa Ferreira, Joao Paulo Carvalho, Sofia Oliveira, Lu´ısa Coheur, Paula Paulino, AM Veiga Sim˜ao, and Isabel Trancoso. Automatic cyberbullying detection: A systematic review. Computers in Human Behavior , 93:333–345,

    work page 2019

  38. [38]

    Approaches to automated detection of cyber- bullying: A survey

    [Salawu et al., 2017] Semiu Salawu, Yulan He, and Joanna Lumsden. Approaches to automated detection of cyber- bullying: A survey. IEEE Transactions on Affective Com- puting, 11(1):3–24,

    work page 2017

  39. [39]

    Emotional and behavioural problems in the context of cyberbullying: A longitudinal study among german adolescents

    [Schultze-Krumbholz et al., 2013] Anja Schultze- Krumbholz, Anne J ¨akel, Martin Schultze, and Herbert Scheithauer. Emotional and behavioural problems in the context of cyberbullying: A longitudinal study among german adolescents. In Emotional and Behavioural Difficulties Associated with Bullying and Cyberbullying , pages 102–118. Routledge,

    work page 2013

  40. [40]

    Attention is all you need

    [Vaswani, 2017] A Vaswani. Attention is all you need. Ad- vances in Neural Information Processing Systems,

    work page 2017

  41. [41]

    Sosnet: A graph convolutional network approach to fine-grained cyberbullying detection

    [Wang et al., 2020a] Jason Wang, Kaiqun Fu, and Chang- Tien Lu. Sosnet: A graph convolutional network approach to fine-grained cyberbullying detection. In 2020 IEEE International Conference on Big Data (Big Data) , pages 1699–1708. IEEE,

    work page 2020

  42. [42]

    Autoformer: Decomposition transform- ers with auto-correlation for long-term series forecast- ing

    [Wu et al., 2021] Haixu Wu, Jiehui Xu, Jianmin Wang, and Mingsheng Long. Autoformer: Decomposition transform- ers with auto-correlation for long-term series forecast- ing. Advances in neural information processing systems , 34:22419–22430,

    work page 2021

  43. [43]

    Cyberbullying detection using pre- trained bert model

    [Yadav et al., 2020] Jaideep Yadav, Devesh Kumar, and Dheeraj Chauhan. Cyberbullying detection using pre- trained bert model. In 2020 International Conference on Electronics and Sustainable Communication Systems (ICESC), pages 1096–1100. IEEE,

    work page 2020

  44. [44]

    Sccd: A session-based dataset for chinese cyberbullying detection

    [Yang et al., 2025] Qingpo Yang, Yakai Chen, Zihui Xu, Yu- ming Shang, Sanchuan Guo, and Xi Zhang. Sccd: A session-based dataset for chinese cyberbullying detection. In Proceedings of the 31st International Conference on Computational Linguistics, pages 9533–9545. Association for Computational Linguistics,

    work page 2025

  45. [45]

    Are transformers effective for time series fore- casting? In Proceedings of the AAAI conference on artifi- cial intelligence, volume 37, pages 11121–11128,

    [Zeng et al., 2023] Ailing Zeng, Muxi Chen, Lei Zhang, and Qiang Xu. Are transformers effective for time series fore- casting? In Proceedings of the AAAI conference on artifi- cial intelligence, volume 37, pages 11121–11128,

    work page 2023

  46. [46]

    Informer: Beyond efficient transformer for long sequence time-series forecasting

    [Zhou et al., 2021] Haoyi Zhou, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li, Hui Xiong, and Wan- cai Zhang. Informer: Beyond efficient transformer for long sequence time-series forecasting. In Proceedings of the AAAI conference on artificial intelligence, volume 35, pages 11106–11115,

    work page 2021

  47. [47]

    Fedformer: Fre- quency enhanced decomposed transformer for long-term series forecasting

    [Zhou et al., 2022] Tian Zhou, Ziqing Ma, Qingsong Wen, Xue Wang, Liang Sun, and Rong Jin. Fedformer: Fre- quency enhanced decomposed transformer for long-term series forecasting. In International conference on machine learning, pages 27268–27286. PMLR, 2022

    work page 2022