Cyberbullying affects one in ten children in Ireland with 32% talking to a stranger online every week. What more do more to protect our children online? Researchers from Saudi Arabia are hoping to tackle this growing issue through the power of Deep learning.
According to a latest report, more than one in ten children in Ireland experience cyberbullying.
Cybersafe Ireland’s Annual report states that :
- 68% of 8 – 13-year-olds own a smartphone.
- 70% use social media.
- 32% talk to strangers online every week. (18% every day)
- 62% of teachers are dealing with online safety incidences in the classroom.
Cyberbullying is any form of bullying or harassment using technology. It has become quite prevalent in modern society, mostly through social media, and can be quite harmful. It can also be life-threatening.
Only recently Gardai have issued a warning regarding the “Momo Challenge” which targets young and vulnerable people on social media. “Momo” allegedly requests that the person performs dangerous acts like self-harm.
Many young people have been affected by this in Ireland and overseas. While it is not technically “cyberbullying,” it has brought the topic of online safety to the forefront.
Currently, there are very few specific laws related to cyberbullying as technology surges ahead. This is a major drawback of social media, particularly when it comes to the protection of young people.
So it is obvious that the internet can be a dangerous place and that a child’s online activity must be monitored. Can AI do more to aid in the detection of cyberbullying? Recent research has shown that it is certainly possible…
Is Deep Learning The Key To Cyberbullying Prevention?
Many deep learning approaches identify cyberbullying work by analyzing textual and user features. However, there are limitations to these techniques that significantly reduce their performance.
For example, current approaches do not consider that some user data can be easily fabricated, such as age and date of birth. A research group out of Saudi Arabia, have developed a new approach to detect cyberbullying on Twitter using deep learning called OCDD.
OCDD represents a tweet as a set of word vectors. Compared to the current method which extracts features from tweets and feeds them to a classifier.
“In this way, the semantics of words are preserved, and the feature extraction and selection phases can be eliminated,” the researchers explain in their papaer publsihed on IEEE Explore.
The researchers built their approach on labelled training data and generated word embeddings for individual words using GloVe. GloVe is an unsupervised learning algorithm that can obtain vector representations for words.
These word embeddings are then fed to a convolutional neural network (CNN) to detect whether they could be associated with cyberbullying. CNN algorithms typically consist of an input and output layer, as well as several other layers. Entering these manually can be quite challenging and time-consuming.
Researchers decided to incorporate a metaheuristic optimization algorithm into their model. Facilitating the process by identifying optimal or near optimal values to be used for classification.
“OCDD advances the current state of cyberbullying detection by eliminating the hard task of feature extraction/selection and replacing it with word vectors which capture the semantic of words and CNN which classifies tweets in a more intelligent way than traditional classification algorithms,”Al-Ajlan and Ykhlef write in their paper.
OCDD attained some positive results and researchers now plan to adapt their approach so that it can also analyze text in Arabic.