Cyberbullying detection in social media text based on character‐level convolutional neural network with shortcuts

Abstract

As people spend increasingly more time on social networks, cyberbullying has become a social problem that needs to be solved by machine learning methods. Our research focuses on textual cyberbullying detection because text is the most common form of social media. However, the content information in social media is short, noisy, and unstructured with incorrect spellings and symbols, and this impacts the performance of some traditional machine learning methods based on vocabulary knowledge. For this reason, we propose a Char-CNNS (Character-level Convolutional Neural Network with Shortcuts) model to identify whether the text in social media contains cyberbullying. We use characters as the smallest unit of learning, enabling the model to overcome spelling errors and intentional obfuscation in real-world corpora. Shortcuts are utilized to stitch different levels of features to learn more granular bullying signals, and a focal loss function is adopted to overcome the class imbalance problem. We also provide a new Chinese Weibo comment dataset specifically for cyberbullying detection, and experiments are performed on both the Chinese Weibo dataset and the English Tweet dataset. The experimental results show that our approach is competitive with state-of-the-art techniques on cyberbullying detection task.

Awareness about Cyber bullying on Social Media among Female Students in a Malaysian Public University

“…The implications include depression, embarrassment, anger, humiliation, lack of self-esteem and confidence, loneliness, suicidal ideas, insecurity, harassment, and family issues (Odora, 2015). Cyberbullying is deemed to create negative online reputations for victims, which is able to impact college admissions, employment, and other areas of life, and may even cause more serious and permanent consequences of self-harm and suicide (Lu et al, 2020). Cyberbullying victims may experience depression or worse may commit suicide.…”

Cyberbullying Detection in Social Networks Using Bi-GRU with Self-Attention Mechanism

“…In these two works, the oversampling method was handled for data processing, which led to overfitting of data, in other words, performance claims of the models in these two works had become overestimated. Lu et al [32] proposed a character-level convolutional neural network with shortcuts, their model turned out that char-level features could be learned to overcome spelling errors and intentional obfuscation in online posts. Zhang et al [33] introduced a method combining the one-dimensional CNN and the single GRU network, they experimented on the dataset of the Twitter platform and obtained an increase between 1 and 13% in F1-score.…”

MaLang: A Decentralized Deep Learning Approach for Detecting Abusive Textual Content

“…The major issues with the hybrid models are that the experimental results and evaluations are bound to a single dataset [10,28]. Automatic improper conversation (such as disgusting language and offensive chats) extraction is complicated and time consuming [32,[49][50][51].…”

“…Prior research utilized different data sources for analyzing cyberbullying detection systems and as such, only the reported performance measures are used for comparison. [22] 0.586 NLP [7] 0.950 Logistic Regression [64] 0.905 0.910 0.900 Decision Tree [23] 0.785 SVM [30] 0.780 0.780 LSTM [38] 0.9397 CNN [50] 0.960 Deep Learning Ensemble [46] 0.980 Tolba et al [65] recently performed an independent comparison of various machine learning and deep learning approaches using various encoding schemas for the detection of online harassment (no age limitations specified). Tolba et al's findings report an optimal geometric mean, F1 score, and AUROC of: 0.8107, 0.8107, 0.7612; all for systems that utilized GloVe with either a Bi-LSTM or LSTM based deep learning classifier.…”