CTChristina Thorpe

Christina Thorpe

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Christina Thorpe has a total of 28 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    An explainable voice-based age assurance system for binary age classification using acoustic features and machine learning

    by Malik Awais Khan, Aurelia Power, Peter Corcoran, Christina Thorpe

  • Top Cited Papers
    Open Access

    Protecting Young Users on Social Media: Evaluating the Effectiveness of Content Moderation and Legal Safeguards on Video-Sharing Platforms

    by Fatmaelzahraa Eltaher, Rahul Krishna Gajula, Luis Miralles-Pechuán, Patrick Crotty, Juan Martínez-Otero, Christina Thorpe, Susan Mckeever

    Video-sharing platforms such as TikTok, YouTube, and Instagram implement content moderation policies to reduce the exposure of minors to harmful videos. As video has become the dominant and most immersive form of online content, assessing how effectively these systems protect younger users is increasingly important. This study evaluates the effectiveness of video moderation for different age groups on TikTok, YouTube, and Instagram, based on a focused set of experimental accounts. Accounts were created for simulated users aged 13 and 18, and 3,000 recommended videos were analyzed in two interaction modes: passive scrolling and search-based scrolling. Each video was manually assessed for the severity of the harm using a unified harm classification framework. While low-severity harm was the most prevalent form encountered, the results show that accounts configured as 13-year-olds encountered harmful videos more frequently and rapidly than accounts configured as 18-year-olds. On YouTube, 15% of videos recommended to 13-year-old accounts during passive scrolling were classified as harmful, compared to 8.17% for adult accounts, with exposure occurring within an average of 3:06 minutes. This exposure appeared without user-initiated searches, highlighting weaknesses in algorithmic filtering. Results from our targeted study point to gaps in video moderation systems, suggesting the need for more effective safeguards to better protect minors from harmful online content.


  • Open Access

    Building societal resilience against child grooming: A design thinking approach to understanding the problem and priorities

    by Christina Thorpe, Fiona Jennings, Armin Shams, Lin Yuan Chong, Kamran Mir, Brian Barry, Susan McKeever, Matt Bowden

  • Open Access

    Protecting Young Users on Social Media: Evaluating the Effectiveness of Content Moderation and Legal Safeguards on Video Sharing Platforms

    by Fatmaelzahraa Eltaher, Rahul Krishna Gajula, Luis Miralles-Pechuán, Patrick Crotty, Juan Martínez-Otero, Christina Thorpe, Susan McKeever

    Video-sharing social media platforms, such as TikTok, YouTube, and Instagram, implement content moderation policies aimed at reducing exposure to harmful videos among minor users. As video has become the dominant and most immersive form of online content, understanding how effectively this medium is moderated for younger audiences is urgent. In this study, we evaluated the effectiveness of video moderation for different age groups on three of the main video-sharing platforms: TikTok, YouTube, and Instagram. We created experimental accounts for the children assigned ages 13 and 18. Using these accounts, we evaluated 3,000 videos served up by the social media platforms, in passive scrolling and search modes, recording the frequency and speed at which harmful videos were encountered. Each video was manually assessed for level and type of harm, using definitions from a unified framework of harmful content. The results show that for passive scrolling or search-based scrolling, accounts assigned to the age 13 group encountered videos that were deemed harmful, more frequently and quickly than those assigned to the age 18 group. On YouTube, 15\% of recommended videos to 13-year-old accounts during passive scrolling were assessed as harmful, compared to 8.17\% for 18-year-old accounts. On YouTube, videos labelled as harmful appeared within an average of 3:06 minutes of passive scrolling for the younger age group. Exposure occurred without user-initiated searches, indicating weaknesses in the algorithmic filtering systems. These findings point to significant gaps in current video moderation practices by social media platforms. Furthermore, the ease with which underage users can misrepresent their age demonstrates the urgent need for more robust verification methods.


  • Open Access

    Discovering Child Sexual Abuse Material Creators’ Behaviors and Preferences on the Dark Web

    by Vuong M. Ngo, Rahul Gajula, Christina Thorpe, Susan Mckeever

    Background: Producing, distributing or discussing child sexual abuse materials (CSAM) is often committed through the dark web in order to remain hidden from search engines and regular users. Additionally, on the dark web, the CSAM creators employ various techniques to avoid detection and conceal their activities. The large volume of CSAM on the dark web presents a global social problem and poses a significant challenge for helplines, hotlines and law enforcement agencies. Objective: Identifying CSAM discussions on the dark web and uncovering associated metadata insights into characteristics, behaviours and motivation of CSAM creators. Participants and Setting: We have conducted an analysis of more than 353,000 posts generated by 35,400 distinct users and written in 118 different languages across eight dark web forums in 2022. Out of these, approximately 221,000 posts were written in English and contributed by around 29,500 unique users. Method: We propose a CSAM detection intelligence system. The system uses a manually labelled dataset to train, evaluate and select an efficient CSAM classification model. Once we identify CSAM creators and victims through CSAM posts on the dark web, we proceed to analyze, visualize and uncover information concerning the behaviors of CSAM creators and victims. Result: The CSAM classifier, based on Support Vector Machine model, exhibited good performance, achieving the highest precision of 92.3\%, accuracy of 87.6\% and recall of 84.2\%. Its prediction time is fast, taking only 0.3 milliseconds to process a single post on our laptop. While, the Naive Bayes combination is the best in term of recall, achieving 89\%, and its prediction time is just 0.1 microseconds per post. Across the eight forums in 2022, our Support Vector Machine model detected around 63,000 English CSAM posts and identified near 10,500 English CSAM creators. The analysis of metadata of CSAM posts revealed meaningful information about CSAM creators and their victims, such as: (1) the ages and nationalities of the victims typically mentioned by CSAM creators, (2) forum topics where the CSAM creators assign their posts, and (3) online platforms preferred by CSAM creators for sharing or uploading CSAM. Conclusion: Our CSAM detection system exhibits high performance in precision, recall, and accuracy in real-time when classifying CSAM and non-CSAM posts. Additionally, it can extract and visualize valuable and unique insights about CSAM creators and victims by employing advanced statistical methods. These insights prove beneficial to our partners, i.e. national hotlines and child agencies.


  • Open Access

    Robustness of Image-Based Malware Classification Models trained with Generative Adversarial Networks

    by Ciaran Reilly, Stephen O Shaughnessy, Christina Thorpe

    As malware continues to evolve, deep learning models are increasingly used for malware detection and classification, including image-based classification. However, adversarial attacks can be used to perturb images so as to evade detection by these models. This study investigates the effectiveness of training deep learning models with Generative Adversarial Network-generated data to improve their robustness against such attacks. Two image conversion methods, byteplot and space-filling curves, were used to represent the malware samples, and a ResNet-50 architecture was used to train models on the image datasets. The models were then tested against a projected gradient descent attack. It was found that without GAN-generated data, the models’ prediction performance drastically decreased from 93-95% to 4.5% accuracy. However, the addition of adversarial images to the training data almost doubled the accuracy of the models. This study highlights the potential benefits of incorporating GAN-generated data in the training of deep learning models to improve their robustness against adversarial attacks.


  • Open Access

    Investigation, Detection and Prevention of Online Child Sexual Abuse Materials: A Comprehensive Survey

    by Vuong M. Ngo, Cach N. Dang, Christina Thorpe, Susan Mckeever

    Child sexual abuse inflicts lifelong devastating consequences for victims and is an ongoing social concern. In most countries, child sexual abuse material (CSAM) distribution is illegal. As a result, there are many research papers in the literature which proposed technologies to detect and investigate CSAM. In this survey, a comprehensive search of the peer-reviewed journal and conference paper databases (including preprints) is conducted to identify high-quality literature. We use the PRISMA methodology to refine our search space to 2,761 papers published by Springer, Elsevier, IEEE and ACM. After iterative reviews of title, abstract and full text for relevance to our topics, 43 papers are included for full review. Our paper provides a comprehensive synthesis of the tasks of the current research and how the papers use techniques and datasets to solve their tasks and evaluate their models. To the best of our knowledge, we are the first to focus exclusively on online CSAM detection and prevention with no geographic boundaries, and the first survey to review papers published after 2018. It can be used by researchers to identify gaps in knowledge and relevant publicly available datasets that may be useful for their research.


  • Open Access

    An ISO/IEC 7816-4 Application Layer Approach to Mitigate Relay Attacks on Near Field Communication

    by Christina Thorpe, John Tobin, Liam Murphy

    Near Field Communication (NFC) has become prevalent in access control and contactless payment systems, however, there is evidence in the literature to suggest that the technology possesses numerous vulnerabilities. Contactless bank cards are becoming commonplace in society; while there are many benefits from the use of contactless payments, there are also security issues present that could be exploited by a malicious third party. The inherently short operating distance of NFC (typically about 4 cm) is often relied upon as a means of ensuring intentional interaction on the user’s part and limiting attack vectors. However, NFC is particularly sensitive to relay attacks, which entirely negate the security usefulness of the short-range aspect of technology. The aim of this article is to demonstrate how standard hardware can be used to exploit the technology to carry out a relay attack. Considering the risk that relay attacks pose, a countermeasure is proposed to mitigate this threat. Our countermeasure yields a 100% detection rate in experiments undertaken – in which over 10,000 contactless transactions were carried out on a range of different contactless cards and devices. In these experiments, there was a false positive rate of 0.38% – 0.86%. As little as 1 in every 250 transactions were falsely classified as being the subject of a relay attack and so the user experience was not significantly impacted. With our countermeasure implemented, transaction time was lengthened by only 0.22 seconds.


  • Open Access

    FedORAM: A Federated Oblivious RAM Scheme

    by Alexandre Pujol, Liam Murphy, Christina Thorpe

    Instant messaging (IM) applications, even with end-to-end encryption enabled, pose privacy issues due to metadata and pattern leakage. Our goal is to develop a model for a privacy preserving IM application, by designing an IM application that focuses on hiding metadata and discussion patterns. To solve the issue of privacy preservation through the obfuscation of metadata, cryptographic constructions like Oblivious Random Access Machines (ORAM) have been proposed in recent years. However, although they completely hide the user access patterns, they incur high computational costs, often resulting in excessively slow performance in practice. We propose a new federated model, FedORAM, which is the first ORAM scheme that uses a federation of servers to hide metadata for an IM use case. In order to investigate the trade-off between security and performance, we propose two versions of FedORAM:Weak FedORAM and Strong FedORAM. Strong FedORAM uses a tree-based federation architecture to ensure strong obliviousness, but with an increased overhead cost. Weak FedORAM has a more simple federated architecture that only uses Oblivious Transfer (OT) to increase communication speed, but with security consequences. Our results show that both constructions are faster than a similar client-server ORAM scheme. Furthermore, Weak FedORAM has a response time of less than 2 seconds per message for a middle-sized federation.