Improving Cyberbullying Detection with User Context

Abstract

Abstract. The negative consequences of cyberbullying are becoming more alarming every day and technical solutions that allow for taking appropriate action by means of automated detection are still very limited. Up until now, studies on cyberbullying detection have focused on individual comments only, disregarding context such as users' characteristics and profile information. In this paper we show that taking user context into account improves the detection of cyberbullying. IntroductionMore and more teenagers in online communities are exposed to and harmed by cyberbullying. Studies 1 show that in Europe about 18% of the children have been involved in cyberbullying, leading to severe depressions and even suicide attempts. Cyberbullying is defined as an aggressive, intentional act carried out by a group or individual, using electronic forms of contact repeatedly or over time, against a victim who cannot easily defend him-or herself [1]. Besides social measures, technical solutions have to be found to deal with this social problem. At present social network platforms rely on users alerting network moderators who in turn may remove bullying comments. The potential for alerting moderators can be improved by automatically detecting such comments allowing a moderator to act faster. Studies on automatic cyberbullying detection are few and typically limited to the individual comments and do not take context into account [2][3]. In this study we show that taking user context, such as a user's comments history and user characteristics [4], into account can improve the performance of detection tools for cyberbullying incidents considerably. We approach cyberbullying detection as a supervised classification task for which we investigated three incremental feature sets. In the next sections the experimental setup and results will be described, followed by a discussion of related work and conclusions.1 EU COST Action IS0801on Cyberbullying (https://sites.google.com/site/costis0801/). 694M. Dadvar et al. Experiment CorpusYouTube is the world's largest user-generated content site and its broad scope in terms of audience, videos, and users' comments make it a platform that is eligible for bullying and therefore an appropriate platform for collecting datasets for cyberbullying studies. As no cyberbullying dataset was publicly available, we collected a dataset of comments on YouTube movies. To cover a variety of topics, we collected the comments from the top 3 videos in the different categories found in YouTube. For each comment the user id, its date and time were also stored. Only the users with public profiles (78%) were kept. The final dataset consists of 4626 comments from 3858 distinct users. The comments were manually labelled as bullying (9.7%) and non-bullying based on the definition of cyberbullying in this study (inter-annotator agreement 93%). For each user we collected the comment history, consisting of up to 6 months of comments, on average 54 comments per user. Feature Space DesignThe following three feature sets were...

Proceedings of the 6th Workshop on Balto-Slavic Natural Language Processing

“…Obscenity and profanity filtering can be seen as a part of developing content filters (such as parental controls (Weir and Duta, 2012)), cyberbullying detectors (Dadvar et al, 2013) and spam filters (Yoon et al, 2010). Another application of obscenity filtering is found in sentiment analysis, where obscene words are treated as indicators of negative (Ji et al, 2013) or sarcastic reviews (Bamman and Smith, 2015).…”

Proceedings of the 17th Annual Meeting of the Special Interest Group on Discourse and Dialogue

“…Prior work in abusive language has been rather diffuse as researchers have focused on different aspects ranging from profanity detection to hate speech detection (Warner and Hirschberg, 2012) to cyberbullying (Dadvar et al, 2013) and to abusive language in general (Chen et al., 2012;Djuric et al, 2015b).…”

“…Token n-grams are one of the most popular features across many works (Yin et al, 2009; Chen et al, 2012;Warner and Hirschberg, 2012;Xiang et al, 2012;Dadvar et al, 2013). Hand-crafted regular expressions and blacklists also feature prominently in (Yin et al, 2009;Xiang et al, 2012).…”

“…Other features and methodologies have also been found useful. For example, Dadvar et al (2013) found that in the task of identifying cyberbullying in YouTube comments, a small performance improvement could be gained by including features which model the user's past behavior. Xiang et al (2012) tackled detecting offensive tweets via semi-supervised LDA approach.…”

Experts and Machines against Bullies: A Hybrid Approach to Detect Cyberbullies

“…Many social studies have been conducted to provide support and training for adults and teenagers [2,3]. The majority of the existing technical studies on cyberbullying have concentrated on the detection of bullying or harassing comments [4][5][6], while there is hardly work on the more challenging task of detecting cyberbullies and studies for this area of research are largely missing. There are few exceptions however, that point out an interesting direction for the incorporation of user information in detecting offensive contents, but more advanced user information or personal characteristics such as writing style or possible network activities has not been included in these studies [7,8].…”