Modeling the Detection of Textual Cyberbullying

Abstract

The scourge of cyberbullying has assumed alarming proportions with an ever-increasing number of adolescents admitting to having dealt with it either as a victim or as a bystander. Anonymity and the lack of meaningful supervision in the electronic medium are two factors that have exacerbated this social menace. Comments or posts involving sensitive topics that are personal to an individual are more likely to be internalized by a victim, often resulting in tragic outcomes. We decompose the overall detection problem into detection of sensitive topics, lending itself into text classification sub-problems. We experiment with a corpus of 4500 YouTube comments, applying a range of binary and multiclass classifiers. We find that binary classifiers for individual labels outperform multiclass classifiers. Our findings show that the detection of textual cyberbullying can be tackled by building individual topic-sensitive classifiers.

Mastodon Content Warnings: Inappropriate Contents in a Microblogging Platform

“…It is becoming increasingly clear that these platforms open space for contents and conversations which often become inappropriate and hurtful towards the online communities that these systems are supporting. This negative phenomenon is further supported by a recent statement by the Facebook VP of product management saying that the Facebook community "submit tens of millions of reports a week about potentially objectionable contents" 1 ; and it is becoming more urgent since children and adolescents easily access to social media but they are more susceptible to these contents and more exposed to threats such as cyberbullying (Dinakar, Reichart, and Lieberman 2011), online harassment or sexting (O'Keeffe, Clarke-Pearson, and others 2011).…”

You Don't Know How I Feel: Insider-Outsider Perspective Gaps in Cyberbullying Risk Detection

“…Given the large-scale datasets required for robust analysis, this task can prove to be labor-intensive and daunting. Consequently, many researchers resort to hiring a group of annotators (Schenk, Guittard et al 2009), or use crowdsourcing tactics (Hosseinmardi et al 2015;Rafiq et al 2015) to manually code the data based on a pre-defined codebook that attempts to describe the phenomena of interest (Dinakar, Reichart, and Lieberman 2011;Dadvar et al 2013;Singh, Ghosh, and Jose 2017). This general practice may be appropriate where the classification task is straightforward, such as identifying specific objects in images.…”

“…Yet, most published cyberbullying detection systems continue to utilize the practice of using third-party ("outsider") annotators in their research. For instance, Dinakar et al investigated the use of classification to detect cyberbullying comments on YouTube (Dinakar, Reichart, and Lieberman 2011). They had two annotators, one of which was a youth educator, label data related to sexuality, race, and intelligence.…”

“…Dadvar et al followed a similar approach to collect and label 4.6k YouTube comments (Dadvar et al 2013). Other studies on cyberbullying detection (Dinakar, Reichart, and Lieberman 2011;Huang, Singh, and Atrey 2014;Singh, Ghosh, and Jose 2017) report using students or other individuals as third-party annotators, but often do not report on the past experiences of these individuals, their positionality to the dataset, or how they were trained to accurately complete the annotation task.…”

You Too?! Mixed-Initiative LDA Story Matching to Help Teens in Distress

“…In the field of computational linguistics, related work for the detection of textual cyberbullying has involved the use of statistical supervised machine learning topic classification to detect sensitive topics such as sexuality, race and culture, socio-normative conflicts, physical appearance and intelligence [5]. The Genesis project has investigated the understanding of automatic plot…”