DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
How to … co-create research with medical students
No full textCo-creation of medical student research projects by both supervisors and students has the potential to enhance student learning and project outcomes. However, it also presents challenges to research supervisors, who need to balance the needs of the project and the student, as well as adapt to the varied motivations and skill levels of students. Similarly, students need to take greater ownership of their projects and actively contribute to decision-making and generation of new ideas. In this new paradigm, traditional hierarchies must be reconceptualised to make space for increased student empowerment, engagement and exploration. This "How to …" paper offers strategies for co-creating research with medical students and highlights potential new roles for students in a more collaborative and equitable supervisor-student relationship
States of reality: how democracy levels impact the framing of deepfakes in newspapers and the resultant impact on attitudes
No full textThis study investigates how the framing of deepfake technology under different government systems affects public perception and actions. Drawing on media framing theory and political communication literature, content analysis of news articles was employed to identify salient frames and examine linguistic differences. Existing literature on framing of emerging technologies has predominantly situated within Western perspectives, underscoring the focus of this research on democratic and non-democratic countries across Asia. The research explores how different countries frame risks and benefits, moral evaluation and proposed prognostic actions. Results indicate that democratic states tend to focus on risks associated with deepfakes, while non-democratic states leaned towards a more balanced approach, highlighting benefits as well. Subsequently, an experimental survey was conducted to measure the differential impact of these framing strategies on public perception and actions. Results revealed that negative framing of deepfakes led to higher levels of concern, lower support for development and higher support for regulations compared to neutral and positive framing. However, there were no immediate effects on detection abilities. The findings provide insights into the interplay between political systems, media narratives and public perceptions of deepfakes as an emerging AI-driven technology and contribute to growing literature surrounding factors impacting deepfake detection abilities.Bachelor's degre
AIView: helping students prepare for software engineering technical interviews using large language models
No full textComputer Science students often find themselves unprepared for Software Engineering interviews as these skills are not sufficiently tested in most universities worldwide. While Data Structures and Algorithms classes help, students face difficulties practicing and honing their
interview skills.
Increase in reasoning skills for generative AI allows for students to potentially better prepare for internship interviews by speaking to a simulated interviewer powered by an LLM. This project aims to create that interview tool which helps students mimic real-life interview scenarios using Generative AI and simulate the interview process. The project also seeks to provide structured feedback to the user aimed to help them improve their overall performance,
ensuring that they're well suited to tackle real-life software engineering interviews.Bachelor's degre
“愤怒即力量”：韩国流行音乐文化中女性愤怒情绪的符号化展演——以BIBI作品为例 = Anger as power: the symbolic performance of female anger in South Korean popular music culture — a case study of BIBI's works
No full text在新媒体时代，韩国流行音乐（K-pop）作为全球流行文化的重要载体，其性别角色建构备受关注。传统女偶像形象长期受父权制影响，被框限于清纯或性感的二元框架中。随着全球女权主义浪潮的兴起，Girl crush风格逐渐成为K-pop偶像突破性别刻板印象的重要媒介，通过自信、独立甚至暴力的形象塑造，挑战传统性别规范，并传达女性愤怒作为面对压迫的正当情绪这一正确认知。本文以女歌手BIBI的音乐作品为个案，分析其三部暴力叙事MV《Animal Farm》、《Vengeance》、《Jotto》，探讨女性愤怒的表达及其象征意义，重点分析女性愤怒如何通过暴力符号实现视觉化表达，并使暴力成为对父权压迫的象征性反抗。研究发现，其作品透过视觉隐喻和叙事策略的创新，使女性愤怒被赋予正当性，成为反抗压迫的工具。研究也强调，文本中的暴力并非对现实暴力的鼓吹，而是一种象征性的反抗方式，为观众提供情绪宣泄的空间，并促使其反思现实中的性别压迫，进而推动现实中的性别平等实践。
In the new media era, K-pop has become a key vehicle of global pop culture, drawing attention to its construction of gender roles. Traditionally, female idol images have been shaped under patriarchy and confined to the binary of "innocence" or "sexiness". With the rise of global feminism, the "Girl Crush" style enables K-pop idols to challenge gender norms by portraying confident, independent, and even violent female figures, conveying the rightful recognition that female anger is a legitimate emotion in response to oppression. Using BIBI’s music videos, Animal Farm, Vengeance, and Jotto as case studies, this paper examines how female anger is visualized through violence and transform the violence into a symbolic act of resistance against patriarchy. Through innovative visual metaphors and narrative strategies, BIBI’s works legitimize female anger as a tool of resistance against oppression. The study further emphasizes that textual violence does not promote real-world violence but serves as a symbolic form of protest, offering emotional catharsis while encouraging critical reflection on gender oppression and ultimately fostering real-world gender equality.Bachelor's degre
Leveraging covariate space for groupings
No full textThis report investigated the imputation methodology that combines Generalized Additive Mixed Models (GAMM) and k-Nearest Neighbours (k-NN) algorithm used previously in the reference study by Nguyen et al. (2023), and further refined it to improve its efficiency based on RMSE values.
Originally designed to estimate country-level trends in treatment-seeking behaviours, Nguyen's method employed Principal Components Analysis (PCA) on covariates to identify most similar country to those missing data with k-NN. This allows imputation of missing data by borrowing random intercepts from similar country to enhance GAMM's predictive accuracy.
Recognising the importance of predictive accuracy and model reliability, this report systematically evaluated the original methodology, proposed significant improvements through multi-neighbour approach and compared with other existing data imputation researched. Simulated datasets with hierarchical structures and temporal dynamics were generated and used to assess model performance.
Testing Nguyen's original model showed high accuracy but also vulnerabilities associated with it. We transitioned to a multi-neighbour method, assigning fractional weights to the multiple nearest-neighbours identified via k-NN to enhance predictive robustness and reduce sensitivity to outliers. By balancing accuracy with stability, an average RMSE improvement of 21.49% was achieved.
Further comparative analysis against traditional imputation methodologies (k-NNI, MICE, CART, PMM, Linear Regression, EM and MI) also showed that the GAMM-k-NN multi-neighbour model's exceptional predictive accuracy and robustness.Bachelor's degre
Interactive sensemaking with SurveySense: enhancing survey insights through human-AI collaboration on a LLM-based platform
No full textSurveys are essential in various fields to gather feedback and information. Efficient data processing is thus crucial for quick, informed decision-making. As the volume of data continues to grow, automation in survey analysis has become increasingly important. This paper proposes a new software tool, SurveySense, which aims to streamline the survey analysis process. A study involving 26 participants was also conducted to evaluate the tool across six dimensions - Insight Quality, Efficiency, Consistency, User satisfaction and Trustworthiness, Human-AI Collaboration and Usability and Interaction. Results showed that SurveySense would be a helpful assistant for survey analysis but human inputs remain crucial in the process to further refine the AI-generated insights to align with personal goals.Bachelor's degre
Phase transitions in coherent and incoherent light walks with imaginary AAH potentials
No full textThe scattering of waves is ubiquitous throughout nature. Depending on the type of crystals and level of disorders, such scattering behaviours can be categorised into three primary behaviours: diffusion, ballistic transport, and localization. Among these, localisation, and more specifically Anderson localisation [1], has drawn considerable attention. Anderson localisation involves a phase transition from delocalisation to localisation, typically observed in systems with spatial or temporal disorder. One model frequently used to explore this phenomenon is the Aubry-Andr´e-Harper (AAH) model [2, 4]. While much research has focused on Anderson localization in Hermitian systems, introducing non-Hermiticity into the system brings additional complexity. This can shift the transition points and alter the overall dynamics of the system. This research aims to delve into Anderson localisation within non-Hermitian photonic crystal lattices, particularly in the presence of temporal disorder, to better understand the effects of non-Hermiticity in such systems.Bachelor's degre
Ask codey: AI tutor for programming education
No full textWhen practising competitive programming style questions online, students encounter significant
challenges when seeking assistance. Human coaches provide high-quality, incremental, and
personalized guidance, making them the most effective form of support. However, they are
difficult to schedule, leading to slow feedback loops, and they come at a high cost, making them
inaccessible to many. Online platforms and forums address these concerns but lack
personalization. Since online resources do not always understand the specific problem a
student is working on, the guidance provided is often generic. AI chatbots provide some
personalization but are often cumbersome, requiring students to manually input code and
explain their problem. Furthermore, most are not optimized for incremental guidance, frequently
offering verbose responses or revealing full solutions which is not conducive for learning. To
solve these challenges we introduce Ask Codey, a web based AI powered competitive
programming learning platform designed to offer proactive, coach-like feedback. Ask Codey
seamlessly blends traditional coding functionalities—such as accessing diverse coding
challenges, compiling and running code against test cases, and debugging—with innovative
AI-driven features. These include a dynamic hint generation module, proactive feedback, and an
interactive LLM chat assistant. Together, these elements harness the creative potential of LLMs
in a controlled manner, unlocking a more effective, engaging approach to learning. In this report,
we detail the motivation, design, and evaluation of Ask Codey, highlighting the research and
engineering insights that ensure our AI assistants are not only useful to end-users but also
cost-effective and fast. Ask Codey currently focuses on K-through-12 competitive programming.Bachelor's degre
Application of generative AI in healthcare
No full textGraph databases are gaining popularity due to their ability to efficiently and intuitively
represent relationships and structures. This project investigates the construction of an
end-to-end pipeline that leverages a Neo4j-based biomedical Knowledge Graph,
integrated with data from SNOMED CT Ontology and NHANES, to support natural
language question answering in clinical and population health domains. The system
comprises of five key components: concept extraction and normalisation using LLM
guided synonym resolution; Cypher query generation via prompt-engineered LLM;
structured querying over Neo4j; agentic evaluation using LangChain to assess
response completeness; and synthesis of user-facing explanations. Meta’s LLaMA-3
8b model was used for the entirety of the project. Findings show that the system is able
to effectively handle simple queries, but struggles at more complex ones. Results also
underscore the importance of hybrid reasoning architectures to combine symbolic
precision and semantic flexibility for effective natural language access to complex
medical data. The developed system therefore is a great starting point to enhanced
integration with medical graph databases, and provides a basis for the integration of
further and multiple database technologies and LLMs.
The full implementation, including setup instructions and sample queries, is available
at: https://github.com/jtan573/FYP_CCDS24-0108/tree/main.Bachelor's degre
Cryptography in a secure chat application
No full textIn today's digital age, secure communication has become a critical concern as users
demand privacy and data integrity across online platforms. This project explores the
application of cryptography in designing a secure chat application that ensures
confidentiality, authentication, integrity, and availability of communication. By
leveraging modern encryption techniques such as end-to-end encryption, asymmetric
key exchanges, and secure hashing algorithms, the proposed chat application
prevents unauthorized access, eavesdropping, and tampering of messages. The
project evaluates existing cryptographic protocols, such as RSA and AES, and
implements a hybrid model to achieve optimal security and performance.
Additionally, practical challenges such as key management, scalability, and user
experience are addressed. The implementation is tested rigorously against simulated
threats, demonstrating its effectiveness in providing a secure and user-friendly
communication platform. This research contributes to the growing need for secure
digital solutions, paving the way for robust applications in personal and enterprise
environments.Bachelor's degre