CRII: CHS: Cyberbullying Detection Using Content and Social Network Analysis

本项目旨在整合社会科学和计算机科学的相关研究，定义自动检测网络欺凌的新方法。网络欺凌是一个严重的社会问题，发生在技术基础上。根据国家犯罪预防委员会最近的一份报告，超过40%的美国青少年报告称受到网络欺凌。这尤其令人担忧，因为多项研究报告称，网络欺凌的受害者往往患有精神和心身疾病。具体而言，本研究将通过对这些欺凌信息交换的社会关系给予应有的关注，推动网络欺凌检测超越文本分析的最新技术。更高的检测准确性将有助于更好地减轻网络欺凌现象，并可能有助于改善每年遭受网络欺凌的数千名受害者的生活。这项研究的结果也将为未来使用社会干预机制来帮助预防网络欺凌事件打开大门。这项研究的结果还将验证和完善现有的网络欺凌理论，并有可能通过创建一波数据驱动的现象分析来推动这一领域的发展。生成的数据集将提供给更大的研究界，从而实现有助于解决这一社会问题的新发现。本研究将定义网络欺凌自动检测的新方法，并验证和完善与网络欺凌相关的社会科学理论。为了理解网络欺凌，社会科学专家们把重点放在了人格、社会关系以及涉及欺凌者和受害者的心理因素上。最近，计算机科学研究人员还开发了通过文本挖掘网络对话来识别网络欺凌信息的自动化方法。然而，只关注文本内容可能会导致对现象的零碎理解和有限的检测性能。因此，本研究探讨：(1)分析网络周围的社会网络特征是否可以提高网络欺凌检测的准确性；(2)通过调查、人种志和基于访谈的方法获得的网络欺凌社会科学研究结果，在更大规模的自动化数据分析中是否成立。通过分析用户之间的社会关系图，并得出朋友数量、网络嵌入性和关系中心性等特征，该项目将验证（并可能改进）社会科学文献中的多种理论，并吸收这些发现，以创建更好的网络欺凌探测器。该项目将产生新的、全面的模型和算法，可用于自动设置中的网络欺凌检测。

This project aims to define new approaches for automatic detection of cyberbullying by integrating the relevant research in social sciences and computer science. Cyberbullying is a critical social problem that occurs over a technical substrate. According to a recent National Crime Prevention Council report, more than 40% of teenagers in the US have reported being cyberbullied. This is especially worrying as the multiple studies have reported that the victims of cyberbullying often deal with psychiatric and psychosomatic disorders. Specifically, this research will advance the state of the art in cyberbullying detection beyond textual analysis by also giving due attention to the social relationships in which these bullying messages are exchanged. A higher accuracy at detection would allow for better mitigation of the cyberbullying phenomenon and may help improve the lives of thousands of victims who are cyberbullied each year. The results of this research will also open doors to employing social intervention mechanisms to help prevent cyberbullying incidents in future. The findings from this research will also validate and refine existing theories on cyberbullying and potentially advance the field by creating a wave of data-driven analysis of the phenomenon. The generated data set will be made available to the larger research community, thus enabling new findings that can help counter this social problem. This research will define new approaches for automatic detection of cyberbullying and validate and refine social science theories related to cyberbullying. To understand cyberbullying, experts in social science have focused on personality, social relationships, and psychological factors involving both the bully and the victim. Recently computer science researchers have also developed automated methods to identify cyberbullying messages by text mining cyber conversations. However, focusing only on the textual content may lead to a piecemeal understanding of the phenomenon and a limited detection performance. Hence, this research investigates: (1) whether analyzing social network features surrounding the network can improve the accuracy of cyberbullying detection, and (2) whether the findings of social science research on cyberbullying obtained via surveys, ethnography, and interview-based methods hold true when tested via automated data analysis undertaken at a much bigger scale. By analyzing the social relationship graph between users and deriving features such as number of friends, network embeddedness, and relationship centrality, the project will validate (and potentially refine) multiple theories in social science literature and assimilate those findings to create better cyberbullying detectors. The project will yield new, comprehensive models and algorithms that can be used for cyberbullying detection in automated settings.