Detecting Discourse-Independent Negated Forms of Public Textual Cyberbullying

Abstract

Cyberbullying is a risk associated with the online safety of young people and, in this paper, we address one of its most common implicit forms -negation-based forms. We first describe the role of negation in public textual cyberbullying interaction and identify the cyberbullying constructions that characterise these forms. We then formulate the overall detection mechanism which captures the three necessary and sufficient elements of public textual cyberbullying -the personal marker, the dysphemistic element, and the link between them. Finally, we design rules to detect both overt and covert negation-based forms, and measure their effectiveness using a development dataset, as well as a novel test dataset, across several metrics: accuracy, precision, recall, and the F1-measure. The results indicate that the rules we designed closely resemble the performance of human annotators across all measures.

The Role of Previous Discourse in Identifying Public Textual Cyberbullying

“…We base the present investigation on the same view that we expressed previously (Power et al 2017;2018) that the presence of explicit terms/expressions does not suffice for a message or post to be classified as public textual cyberbullying; it must be linked to or it must target a specific person, or group of people. Thus, an instance constitutes public textual cyberbullying if it contains (either explicitly or implicitly) the personal marker/pointer -which identifies or points to the victim(s), the dysphemistic element -which is defined by Allan and Burridge (2006, 31) as the "word or phrase with connotations that are offensive either about the denotatum and/or to people addressed or overhearing the utterance", and the link between the previous two elements -which captures how the dysphemistic element targets the victim(s) identified or pointed to by the personal marker.…”

“…Even those studies that target implicit forms of cyberbullying (Chen et al 2012;Dinakar et al 2012;Nitta et al 2013;Ptaszynski et al 2010;Ptaszynski et al 2016) do not clearly define the boundaries of what is cyberbullying and the approaches described result in a considerable amount of false positives 1 that contain rude and violent language, despite the fact that the use of this type of language does not constitute cyberbullying on its own. On the other hand, although the focus of previous research in the field of cyberbullying detection has been to reduce the number of false negatives 2 , no other study has investigated the linguistic role of prior discourse in identifying public textual cyberbullying, and the purpose of the present paper is to determine the contribution of antecedent messages in resolving the missing cyberbullying elements that we previously proposed (Power et al 2017;2018): the personal marker, the dysphemistic element, and the cyberbullying link between the personal marker and the dysphemistic element.…”

“…To analyse discourse-dependent instances, we used the same dataset as in previous work to which the same cleaning and pre-processing techniques were applied (Power et al 2018). The dataset was acquired from two sources: from Kavanagh' research (2014) on cyberbullying detection and from Hosseinmardi et al (2014a) and Hosseinmardi et al (2014b) dataset from which we randomly selected a continuous portion.…”

“…Once all the cyberbullying elements are inferred from previous discourse and the cyberbullying form identified, the same detection mechanism that we described in previous work (Power et al 2018) can be applied to any instance. However, to identify the missing elements we first need to consider the types of grammatical relations that represent them and because the cyberbullying link cannot appear on its own in discourse dependent instances, but it must be accompanied at least by its grammatical subject, there are only four discourse dependent cyberbullying constructions that we need to discuss: (1) fully inferable constructions -where all three cyberbullying elements, the personal marker, the dysphemistic element, and the link between them, are not explicitly present, but can be inferred from previous discourse, (2) personal marker and cyberbullying link inferable constructions -where the dysphemistic element is explicitly present, but the personal marker and the link must be inferred from previous discourse, (3) dysphemistic element and cyberbullying link inferable constructions -where the personal marker is explicitly present, but the dysphemistic element and the cyberbullying link are entities inferable from previous discourse, and (4) dysphemistic element inferable constructions -where the personal marker and the link are explicitly present, but the dysphemistic element must be inferred from prior discourse.…”

“…In addition, cyberbullying instances must occur in the cyberspace which allows these acts to transcend the physical and temporal constraints that apply to face-to-face bullying. Because of the ability of cyberspace to persist over time, to be searched for repeatedly, to be replicated numerous times, and to multicast to potentially large invisible audiences (Boyd 2007), cyberbullying acts implicitly satisfy the criterion of repetition (Dooley and Cross 2010;Grigg 2010;Langos 2012;Power 2017;2018).…”