An Examination of Regret in Bullying Tweets

Proceedings of NAACL-HLT 2013, pages 697–702, Atlanta, Georgia, 9–14 June 2013. c⃝2013 Association for Computational Linguistics An Examination of Regret in Bullying Tweets Jun-Ming Xu, Benjamin Burchfiel, Xiaojin Zhu Department of Computer Sciences University of Wisconsin-Madison Madison, WI 53706, USA {xujm,burchfie,jerryzhu}@cs.wisc.edu Amy Bellmore Department of Educational Psychology University of Wisconsin-Madison Madison, WI 53706, USA abellmore@wisc.edu Abstract Social media users who post bullying related tweets may later experience regret, potentially causing them to delete their posts. In this pa- per, we construct a corpus of bullying tweets and periodically check the existence of each tweet in order to infer if and when it becomes deleted. We then conduct exploratory analy- sis in order to isolate factors associated with deleted posts. Finally, we propose the con- struction of a regrettable posts predictor to warn users if a tweet might cause regret. 1 Introduction A large body of literature suggests that participants in bullying events, including victims, bullies, and witnesses, are likely to report psychological adjust- ment problems (Jimerson, Swearer, and Espelage, 2010). One potential source of therapy for these is- sues can be self-disclosure of the experience to an adult or friend (Mishna and Alaggia, 2005); exist- ing research suggests that victims who seek advice and help from others report less maladjustment than victims who do not (Shelley and Craig, 2010). Disclosure of bullying experiences through so- cial media may be a particularly effective mecha- nism for participants seeking support because so- cial media has the potential to reach large audi- ences and because participants may feel less inhi- bition when sharing private information in an on- line setting (Walther, 1996). Furthermore, there is evidence that online communication stimulates self- disclosure, which leads to higher quality social rela- tionships and increased well-being (Valkenburg and Peter, 2009). Online disclosure may also present risks for those involved in bullying however, such as re- victimization, embarrassment, and social ostraciza- tion. Evidence exists that some individuals may re- act to these risks retroactively, by deleting their so- cial media posts (Child et al., 2011; Christofides, Muise, and Desmarais, 2009). Several relevant mo- tives have been found to be associated with delet- ing posted information, including conflict manage- ment, safety, fear of retribution, impression manage- ment, and emotional regulation (Child, Haridakis, and Petronio, 2012). Our previous work (Xu et al., 2012) demonstrates that social media can be a valuable data source when studying bullying, and proposes a text categorization method to recognize social media posts describing bullying episodes, bullying traces. To better under- stand, and possibly prevent, user regret after posting bullying related tweets, we collect bullying traces using the same method and perform regular status checks to determine if and when tweets become in- accessible. While a tweet becoming inaccessible does not guarantee it has been deleted, we attempt to leverage http response codes to rule out other com- mon causes of inaccessibility. Speculating that re- gret may be a major cause of deletion, we first con- duct exploratory analysis on this corpus and then re- …

Content truncated to 500 words