[2026] SPARK: Exploiting LLMs and Knowledge Graphs to Build Structured and Personalized Learning Paths (€5,000)
This project (co-supervised with Nadia Yacoubi Ayadi) explores how Large Language Models (LLMs) and knowledge graphs can be combined to develop next-generation intelligent tutoring systems (ITS) that support personalised learning, effort regulation, and goal setting. The core idea is to empower learners to actively select and plan their own learning pathways across the vast landscape of online educational resources.
The work proceeds in two main steps.
- Knowledge extraction and structuring: Using LLMs and NLP tools, the project aims to automatically extract pedagogical elements—learning outcomes, prerequisites, topics, and skills—from diverse course materials and organise them into an Educational Structured Knowledge Graph (EdSKG). This step addresses challenges related to the heterogeneity of course resources, dataset alignment, and the lack of annotated data for concept extraction.
- Personalised learning pathway generation: The EdSKG will then be used to recommend courses and curricula tailored to each learner’s background, goals, and competencies. Leveraging automated planning and explainable AI, the system will provide transparent, visual explanations (e.g., “what-if” scenarios) to help learners understand and control their learning trajectories.
Ultimately, the project aims to create an adaptive, explainable ITS that promotes self-regulated and personalised learning through structured educational knowledge and intelligent recommendation mechanisms.
[2025] CADE-AI - Collaborative Argumentation and Debate Enhancement with AI (€15,000)
This project has two objectives:
(OB1) Enable the automatic mapping of debates into structured format via lightweight LLMs. This objective will require the creation of a multi-task argumentation dataset, the training of a custom LLM model, and its usage on argumentation graphs to assist users.
(OB2) Provide hybrid argumentation semantics to exploit the data of structured debates. The idea is to explore how argument quality dimensions can be used to create new semantics and how they perform in real-life.
[2024] HAGARICE - Harnessing Argumentation Graphs in Augmented Reality for Immersive Co-Creation and Exploration (€15,000)
This project aims to develop the tools for an improved collaborative creation of argumentation graphs and their exploitation for assisted reasoning. Existing online argumentation platforms often depict a debate using a directed graph, offering a visual representation that enhances human grasp of the arguments and their interconnections. These platforms have various applications in domains such as education or e-democracy, where they facilitate broad public involvement in the development of laws. However, there are several problems preventing their adoption for assisted reasoning:
- Information overload - As these platforms become increasingly prevalent, the volume of arguments and options within these systems continues to grow exponentially. This poses a significant hurdle for human analysts, who struggle to manually track changes, understand the various arguments and their relationships, and make sense of the vast amount of information;
- Lack of guidance - Users are not assisted in constructing their various arguments within a debate. This deficiency can result in irrationality, omission of critical information, and the establishment of relationships with incorrect polarity, particularly when confronted with complex arguments;
- Incomplete graphs - Users introduce new support or counter arguments into the system in reaction to a specific local argument. Typically, these new arguments may be redundant or engage with remote arguments that users have not taken into account, resulting in incomplete graphs. Consequently, the conventional argumentation semantics outlined in the literature become inapplicable;
- Asynchronous reasoning - Current research in real-time assisted reasoning within real-life debates using argumentation graphs remains limited. Existing approaches predominantly involve manual argument formalization or users selecting arguments from dropdown menus, interrupting the natural flow of the debate.
This project will provide solutions for the aforementioned problems.
[2024-2026] SMARTER - Structured Multi-sourced AI Reasoning for Enhanced Decision-Making (€6,000)
In this project, we will instantiating a custom large-language model based on the recent Meta AI’s Llama 2 70B model and training it with both synthetic natural language arguments and arguments extracted from existing platforms. We will perform training through LoRA, a novel method to fine-tune LLMs that freezes the pre-trained model weights and adds trainable rank decomposition matrices to each layer of the Transformer architecture, which can capture task-specific features with fewer parameters. This synergy between argumentation graphs, knowledge graphs, and the LLM will be integrated in a decision-support system and evaluated in various decision scenarios.
[2023-2026] Achieving Self-directed Integrated Cancer Aftercare (ASICA) in melanoma: Developing and enhancing the intervention using evidence from the pilot study and emergent technologies (£421,293.28)
This project, funded through Cancer Research UK, is led by Prof. Peter Murchie and co-investigated by myself, Dr. Dewei Yi, and Dr. Julia Allan. This project will optimize the existing platform (ASICA) for integration into the UK National Health Service melanoma survivorship care. ASICA is a theory and evidence-based digital intervention designed to improve the early detection of melanoma recurrences and second primaries by prompting and supporting people with melanoma to conduct regular total skin self-examination. This project has three workpages.
- WP1: Maximising user engagement and adherence
- WP2: Streamlining ASICA to maximize clinical efficiency
- WP3: Simulated interaction between an ASICA user and chatbot
[2023-2024] ZEEFLEET Tool Feasibility Study (£19,929.35)
This project, funded through Innovate UK, is led by Prof. Nir Oren and co-investigated by myself and Dr. Andrew Starkey. With the help of our partner, Better Environment and Transport (BEAT), the goal of this project is to deliver a prototype design and supporting technical analysis to identify which AI and Operations Research techniques best support decisions around the specification and deployment of zero-emission specialist vehicles and recharging infrastructure, as well as routing of vehicles at a realistic (eg. city/authority) scale.
[2022] Verifying Building Code Regulations from Drawings (£5,000)
Currently, the process of ensuring compliance with building regulations demands substantial human effort. Our innovative solution aims to leverage artificial intelligence to revolutionize this process. Through the application of machine learning techniques, our goal is to develop a system capable of 'reading' architectural drawings and employing normative reasoning and argumentation theory to assess their compliance with building regulations. The envisioned tool promises to dramatically reduce the time required for validation, fundamentally transforming the validation process itself. This advancement stands to yield considerable resource savings for architectural firms by expediting the assessment of plans for regulatory adherence. To substantiate our approach, we plan to focus on a specific subset of regulations—initially targeting fire safety regulations. This focused validation will serve as a proof of concept, mitigating the risks associated with the commercialization of this innovative idea.
[2021] Verifying the Compliance of Argumentation Principles with Human Reasoning (£5,000)
The theory of computational argumentation has gained significant attention across various disciplines including computer science, artificial intelligence, psychology, linguistics, and philosophy. It finds application in diverse fields such as medicine, law, agribusiness, and engineering design rationale, where it helps justify decisions made by AI technologies to end-users. These applications rely on specific techniques that adhere to argumentation principles, ensuring rational functioning. However, it remains uncertain whether these principles align with human reasoning and to what extent. By establishing this connection, the gap between human reasoning and formal argumentation can be bridged, leading to improved explanations for end-users and increased trust in AI applications. The project aims to evaluate whether people understand these argumentation principles, if their reasoning aligns with them, and how satisfaction of specific principles affects others, through user experiments and handcrafted examples.
