Machine Learning and Intelligent Communications. 4th International Conference, MLICOM 2019, Nanjing, China, August 24–25, 2019, Proceedings

Research Article

Cyberbullying Detection with BiRNN and Attention Mechanism

Download
169 downloads
  • @INPROCEEDINGS{10.1007/978-3-030-32388-2_52,
        author={Anman Zhang and Bohan Li and Shuo Wan and Kai Wang},
        title={Cyberbullying Detection with BiRNN and Attention Mechanism},
        proceedings={Machine Learning and Intelligent Communications. 4th International Conference, MLICOM 2019, Nanjing, China, August 24--25, 2019, Proceedings},
        proceedings_a={MLICOM},
        year={2019},
        month={10},
        keywords={Attention model Cyberbullying detection Text classification Social network},
        doi={10.1007/978-3-030-32388-2_52}
    }
    
  • Anman Zhang
    Bohan Li
    Shuo Wan
    Kai Wang
    Year: 2019
    Cyberbullying Detection with BiRNN and Attention Mechanism
    MLICOM
    Springer
    DOI: 10.1007/978-3-030-32388-2_52
Anman Zhang1,*, Bohan Li,*, Shuo Wan1, Kai Wang1
  • 1: Nanjing University of Aeronautics and Astronautics
*Contact email: zhanganman@nuaa.edu.cn, bhli@nuaa.edu.cn

Abstract

While the social network has brought a lot of conveniences to our lives, it has also caused a series of severe problems, which include cyberbullying. Cyberbullying is an aggressive and intentional act carried out by a group or an individual to attack a victim on the Internet. Most of the existing works related to cyberbullying detection focus on making use of swearwords to classify text or images with short titles. Although previous methods such as SVM and logistic regression show some advantages in the accuracy of detection, few of them capture the semantic information of non-swearwords which could also make big difference to the final results. In this paper, we propose to use BiRNN and attention mechanism to identify bullies. BiRNN is used to integrate the contextual information, and the attention model reflects the weight of different words for classification. Meanwhile, we convert the severity calculated by the attention layer to the level of cyberbullying. Experiments conducted on three real-world text datasets show that our proposed method outperforms the state-of-art algorithms on text classification and identification effect.