Protect Children Online Safety on Social Media and Mobile Platforms

Open Access
Author:
Chen, Ying
Graduate Program:
Computer Science and Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
June 10, 2014
Committee Members:
  • Sencun Zhu, Dissertation Advisor
  • Heng Xu, Dissertation Advisor
  • Wang Chien Lee, Committee Member
  • Alan Maceachren, Committee Member
Keywords:
  • Children
  • Mobile Apps
  • Mobile Ads
  • Android
  • iOS
  • Social Media
  • Cyberbullying
  • Problematic Content
  • Solicitation
  • Text Mining
  • Privacy
Abstract:
Recent advances in social media and smart mobile technologies make ubiquitous information and communication environments more of a technical reality than a distant vision. The sweeping popularity of these information and communication environments also affects children and adolescents (later referred as “children”). There is a dearth of systematic scholarly work examining those threats for children’s protection. In addressing this void, the goal of our research is to explore and understand the risks and threats of such information and communication environments on children. In particular, our primary research interests lie in the three main threats to children’s online safety (Palfrey et al., 2008): online harassment and cyberbullying, exposure to problematic contents, sexual solicitation and Internet-initiated offline encounters. Online harassment and cyberbullying mostly occur in social media, and always involve offensive language. Since the textual contents on online social media are highly unstructured, informal, and often misspelled, existing research on message-level offensive language detection cannot accurately detect offensive content. Meanwhile, user-level offensiveness detection is an under researched area. We propose the Lexical Syntactic Feature (LSF) architecture to detect offensive contents and identify cyber bullies. Results from experiments showed that the LSF framework achieves accuracy of 96.6%, 77.8% in sentence and user offensiveness detection respectively. Besides, the processing speed of LSF is 10 ms per sentence, suggesting its potential for effective deployment in social media. Problematic contents and solicitation mostly appear on mobile platforms. Since people are heavily relying on the search results and rankings on app stores to explore new apps, on the market side, we must help parents easily choose appropriate apps for their kids with minimal online threats. Since we concern about problematic contents and cyber solicitation, and posting personal information and interacting with online strangers are the two major types of behaviors expose children to cyber solicitation (Ybarra et al., 2007), we monitor two types of risks on mobile devices: content risks and privacy risks. For content risks, we develop automatic mechanisms to detect problematic contents and verify the content ratings of mobile apps and in-app ads. The results show that 27.8% of Android apps have unreliable content ratings. In addition, a large percent of the in-app advertisements carry inappropriate contents for children, and 35.9% of Android in-app ads and 38.9% of iOS in-app ads are found exceeding the host apps’ content ratings. For privacy risks, we adopt the contextual integrity theory (Nissenbaum, 2009) to develop quantitative measures of privacy risks on mobile apps. We also propose an automatic system with user interface to assist parents being aware of apps’ privacy risks, therefore, to make informed decisions when choosing apps for their children. We believe that the findings have important implications for the legal and educational departments, social medias, platform providers (e.g., Google or Apple), as well as for regulatory bodies and application developers.