Projects running on Hex

The following are projects that are currently running (or have been run) on Hex. They range from research staff working on AI, ML and other topics; to PhD students exploring new methods and technologies for NLP (and beyond!).

See also our Publications page for related works.

Projects in 2024
- Detecting LLM Generated Text in Product Reviews
  
  The project will involve creating a dataset of human-written and LLM-generated product reviews. This dataset will be used to train text classification algorithms to determine whether a product review is written by a human or by an LLM.
- Measuring the Effect of Availability Attacks on Dynamic Obstacle Location Data for Reinforcement Learning Agents in 2D Grid World Environments
  
  In a 2D grid world environment, reinforcement learning (RL) agents are trained to navigate toward a goal while avoiding obstacles that move randomly within the environment. However, if an attacker introduces delays in the obstacle location data provided to the agent, it could disrupt the agent's ability to make timely and accurate decisions. This research investigates how such availability attacks affect the agent’s performance and success rate in reaching its goal. The study provides an understanding of the impact of availability attacks in this scenario, contributing to the development of more robust RL systems.
- Investigating How Users Adapt Their Language Across Different Online Social Groups
  
  The proposed project aims to investigate how online users adapt their language across different online communities, focusing on the extent of language mirroring within these environments. Building off research on language mirroring in face-to-face interactions, this project will use Natural Language Processing (NLP) techniques to analyse language patterns in data collected from Reddit.
- Automatic essay scoring
  
  The project aims to provide a new methodology to evaluate automatic essay scoring systems. It also seeks to build a state-of-art language model to score Arabic essays.
- NLP Odyssey
  
  This NLP application helps with Memory-Augmentation for Individuals dealing with Young-Onset Dementia.
- 3rd year project(NLP odessy)
  
  The project aims to explore the potential of Natural Language Processing (NLP) to assist users in summarizing key points from virtual meetings during work hours and lectures. The primary goal is to support individuals with memory impairment while also offering students effective tools to study lecture material.

Projects in 2025
- Spatial Narratives
  
  Extracting and analysing spatial narratives from textual data
- Wmatrix annotation parallelisation (qpym2)
  
  Large scale parallelisation of corpus annotation pipelines
- Applications of LLMs for UK court procedures
  
  Legal NLP has been studied over decades and the recent advancements of the LLMs have pushed the possibilities of practical applications of Legal NLP to greater heights. We study how these LLMs can help lawyers, judges and general public in multiple legal tasks such as Legal judgment prediction, Prior case retrieval and Citation network analysis.
- Various NLP projects
  
  NLP projects, including processing large diachronic datasets, looking at language change over time.
- LM Applications
  
  This project focuses on developing language model (LM)-based applications for natural language processing (NLP) tasks, mainly aiming at the models' learning, multilingual and explainable capabilities.
- Transformer-Assisted LLM Source Code Summarisation
  
  Neural Source Code Summarisation (NSCS) aims to generate natural language summaries of source code to improve developer and maintainer understanding of code. Many solutions to this problem use small transformer models, designed to be run locally on a workstation. Transformer-generated summaries often score well across many NLG metrics but fail to consistently produce clear and understandable natural language. Conversely, Large Language Model (LLM)s’ ability to generate clear and understandable natural language presents an exciting solution to this problem, especially with the increased availability of LLMs and the increase in capability of workstation hardware over recent years meaning that some LLMs can be run from developers' workstations. However, LLM summaries of code often differ greatly from developer-written summaries, and frequently miss key words and phrases resulting in low scores across NLG metrics. We show how combining these two methods by using transformer-generated summaries in prompt engineering may enable LLMs to create better source code summaries.
- Sinhala Encoder-only Language Models
  
  The recent developments of language models (LMs) have shown significant advancements in the field of natural language processing (NLP) as they have produced state-of-the-art results in many NLP tasks, outperforming previous machine learning models such as LSTMs. However, their effectiveness is largely dependent on having access to language resources for model pre-training. This project aims to build LMs for Sinhala; a low-resource language.
- Detecting Fake News in The Era of LLMs
  
  This project investigates the ability of SOTA models to differentiate between real news, human-generated fake news, and LLM-generated fake news.
- Automatic Image Captioning in News Media Articles
  
  Image captioning serves the fundamental goal of creating a descriptive caption for an image. It involves recognising the objects within the image, any actions taking place, their connections, and even subtle details that might not be immediately visible. Once these elements are identified, the task shifts to crafting a concise and pertinent description that accurately captures the essence of the image in terms of syntax and meaning.
- Advancing Sentiment Analysis through Deep Learning Techniques and Language Learning Models
  
  Building DL and LLMs for Sentiment Analysis of Arabic Dialects in Hospitality.
- Machine Translation in the AI Era: Comparing previous methods of machine translation with large language models
  
  Research paper comparing multiple different methods of machine translation and comparing their efficacy, will be comparing rules based, statistical, neural and large language model translation.
- Working Title -- "Cyberbullying on the internet: Comparative study of NLP methods and the role of context".
  
  Comparing the efficacy of rule-based, deep learning, and large language models in determining whether online content is indicative of cyberbullying across five datasets, two languages, with and without additional numerical context.
- Detecting Fake News in The Era of LLMs
  
  This project evaluates the effectiveness of the state-of-the-art (SOTA) detection models in identifying LLM-generated fake news, comparing their accuracy to that of detecting human-generated fake news and real news.
- Spanish Legal NLP
  
  With this project we want to assess the capability of state-of-the-art LLMs to understand actual cases judged by the Spanish Supreme Court. Our main goals are the prediction of the judges decisions and the assessment of the reasoning provided by said LLMs.
- Cross Lingual Summarisation
  
  A Lancaster VinUni collaboration on cross lingual fact-checked summarisation
- A Comparative Study of Vision Transformers and Multimodal Language Models for Violence Detection in Videos
  
  This project compares two artificial intelligence models for detecting violence in videos: TimeSformer, a vision-based model, and LLaMA 3.2 Vision, a multimodal model that combines image and language understanding. It uses public datasets to identify the most effective approach for real-world safety applications.
- Computational Modelling of Energy Materials
  
  The project applies first-principles methods, primarily density functional theory (DFT), to model energy materials such as metal halide perovskites. It focuses on predicting structural, electronic, and defect properties to understand and optimize their performance in photovoltaics and optoelectronics. The goal is to accelerate the design of stable, high-efficiency materials for next-generation energy technologies through atomistic simulations.
- Multilingual Machine Learning Model for PyMUSAS
  
  To create a multilingual machine learning based model that can generate USAS tags, which will then be incorporated into PyMUSAS.
- Davy Notebooks Project
  
  Identify text reuse across Davys Notebooks.
- N8CIR historical spelling internship
  
  Using hex to run the moses smt system to perform character based smt for normalising the spelling of historical text. Also may want to use it for distributed training and testing of other normalisation methods in python.
- GPU Geophysical Inversion for Resistivity Surverys
  
  Accelerating the inversion of large resistivity datasets using parallelisation, GPU offload and surrogate modelling. This will then lead to integration with coupled groundwater models, with application to geohazards (i.e. landslides).
- Enhanced accuracy and efficiency in LLMs through Knowledge Distillation
  
  This project will evaluates different methods of Knowledge Distillation to enhance the factual accuracy and deployment efficiency of Large Language Models.
- Lindenmayer Graphs: An Approach to Rapidly Generate Structured Random Graphs
  
  Network science relies on synthetic data, this project looks at using a type of exponentially parallelizing deterministic context free rewrite system and their application to rapidly generate random graphs with formally guaranteed substructures.
- Detecting LLM generated reviews using ML
  
  My project is going to explore methods in detecting LLM generated reviews, particularly on amazon, using Machine Learning. My current approach will be combining three unique ML models into a ensemble for more accurate classification.
- Analysis of Populist Metaphors for Immigration in Spanish Parliamentary Discourse
  
  This project will develop a computational pipeline for detecting and analysing metaphors in Spanish Parliamentary discourse. A RoBERTa transformer will be fine-tuned using both the VU Metaphor corpus and a manually annotated section from the corpus
- LLM Generated Summaries in Different Forms – Can We Detect Them?
  
  I am refining a pre trained llama model to study how well an ai detector i make that focuses on short summaries and multiple languages can perform.
- FlickType (SCC.402 Group Project)
  
  New implementation for VR typing, moving keyboard to improve the VR typing experience and efficiency has a predictive element requiring AI model
- N8CIR Internship - Student attendance data
  
  Cleaning student attendance data (across iLancaster attendance, Moodle and Panopto) and tracking engagement across the 3 sites and investigating what an "engaged" student looks like and how well they perform.
- Efficient Numerical Integration Schemes for Stochastic Differential Equations
  
  This project explores numerical integration schemes for stochastic differential equations (SDEs) with a particular focus on discretization of ergodic SDEs and applications in Markov Chain Monte Carlo sampling.
- Using Bayesian inference to constrain jet models for a large sample of GRB afterglows
  
  We use Bayesian inference to condition computational models on observations of GRB afterglows to recover the underlying physical parameters.

Projects in 2026
- Evaluation of Cultural Values Represented By Large Language Models in Language Variations
  
  The proposed project will aim to create a benchmark to evaluate large language models’ (LLMs’) representation of the values of various cultures when prompted in languages other than English or the culture’s own language. This will involve creation of a persona-based prompt dataset, which will then be translated into multiple languages and then fed into various LLMs. The output received from the LLMs will be used to evaluate if the LLMs can accurately represent the cultural values (as defined by the Hofstede Culture Model and wave 7 of the World Values Survey) of the cultures queried about.
- Exploring the Ability of AI Chatbots to Offer Empathetic and Supportive Interactions in Mental Health Contexts
  
  The project will aim to investigate AI chatbots and their ability to provide meaningful emotional support to help relieve stress and combat any other mental health issues that an individual may be facing.
- An Explainable Approach to AI-Generated Text Detection for Product Review Moderation
  
  I will be building a text classifier to detect whether a product review is AI generated or human written along with a plain English explanation for the model’s outcome. This will be used by moderation teams to triage faster and make better decisions.
- Multi-layering LLM for multi-task Integration in Offensive text-moderation.
  
  To investigate problems within why a base decoder LLM struggle with hate speech detection and enable a method to mitigate the effects of the problem. To improve the results of the base decoder LLM by giving it additional information through layering other LLM to perform task on classification of the input text. Results includes reasoning and statistical scores such as F1 and accuracy. Analysis of the AI's reason output and how useful it is within the real world though suggestion to moderators of users of a platform.
- Multi-layering LLM for multi-task Integration in Offensive text-moderation
  
  The project proposes to uses many layers of LLM that each is set to classify a piece of text to be inputted within the prompt along with the input text to enable to decoder LLM to understand the context of the message better and generate a reason for the offensiveness classification that matches it. Classifying text on topic, emotion, irony and target can help the final offensive classifier to be more accurate and the reasoning to be more trustworthy and understanding
- AI Enhanced Bid Writing: A Multi-Agent Approach to Tender Automation
  
  This project develops an AI Enhanced Bid Writing system designed to automate responses for complex defense tenders. This project encompaces work for SCC.419 and BAE Systems (FalconWorks).
- A Vision – Language Approach to Automated Mammogram Grading
  
  I am making a vision language model to predict the BIRADs score, density percentage and density classification of breasts from mammogram images.
- A Vision – Language Approach to Automated Mammogram Grading
  
  I am developing a vision language model to predict the BIRADs score, density percentage and density class of breast images from mammograms.
- A Vision – Language Approach to Automated Mammogram Grading
  
  I am developing a vision language model to predict the BIRADs, density percentage and density classification of breast mammogram images.
- CIdiomBench: A Benchmark for Evaluating LLMs on Chinese Idioms
  
  This study aims to develop an evaluation framework to assess whether large language models can accurately comprehend the extended meanings of Chinese idioms. Through Three tasks, i evaluate whether large language models demonstrate a solid understanding of Chinese idioms.
- Context Privacy
  
  This project will develop techniques to limit the information adversaries can obtain from cyber-physical systems.
- Mitigating geohazards through coupled multiphysics modelling and uncertainty analysis
  
  This project aims to improve our ability to predict moisture induced landslides and embankment failures by improved inversion of Electrical Resistivity Tomography. We will start by investigating surrogate models and ensemble methods to quantify uncertainty in ERT data, before propagating this uncertainty into coupled hydrogeophysical models.
- Amazon Intelligence: A Scalable Big Data Pipeline for Analytics and Recommendation Systems
  
  This project architectures a robust, distributed pipeline to ingest and analyze the massive Amazon Reviews 2023 dataset, utilizing advanced database indexing and vector similarity to map complex consumer-product relationships at scale. By deploying scalable machine learning models for clustering and personalized recommendations, the system demonstrates how to transform millions of raw data points into high-fidelity business intelligence and optimized user experiences.
- LARGE SCALE PLATFORMS FOR AI AND DATA ANALYSIS: Amazon Reviews
  
  The project is to create a recommendation system using the amazon reviews and products dataset.
- SCC454 Group E Project
  
  a big data project on the amazon reviews 2023 dataset.
- SCC454 - Large scale systems
  
  As part of the larg scale module we are working with 90GB of data doing clustering and other kind of algorithms. To do on our machines we are using a small subset of data and uni labs are limited to 10GB. Now we have almost compelted code we would like to run on all the data and see fulle final results
- Classifying Political Speeches using Natural Language Processing
  
  Ill be using different NLP methods to attempt to classify US political speeches by party.
- SCC.453 Natural Language Processing - "When Does Text Preprocessing Help or Hurt? A Comparative Study Across Classical Models, Transformers, and LLMs"
  
  I will be designing different preprocessing conditions (like: Strip + Normalisation, Lowercasing, Stopword removal, lemmatisation) and testing different combinations of those preprocessing conditions across classical models, transformer-based models and Large Language Models (These are only a few examples of the preprocessing conditions I plan to design).
  Now, for a good delivery of this project, I would like to run many controlled variants and measure small deltas. Also, I may need to fine-tune transformer models like DistilBERT which involves backpropagation. I will be running multiple experiments on my chosen datasets: Stanford Sentiment Treebank – 2 (SST-2) and TweetEval. I plan to run multiple seeds and multiple conditions to ensure reproducibility of my work.
- Binary Human-Written vs. Machine-Generated Text Classification
  
  Given a full text, determine whether it is human-written or machine-generated. This is based on the Sem Eval 2024 task 8
- Base vs Instruction-Tuned Large Language Models: Evaluation and Calibration of Hallucinations
  
  This project investigates the differences between base and instruction-tuned large language models in terms of hallucination behaviour, evaluating their factual reliability and response calibration. It aims to apply and compare evaluation metrics and calibration techniques to better understand and mitigate hallucinations in generative NLP systems.
- Base vs Instruction-Tuned Large Language Models: Evaluation and Calibration of Hallucinations
  
  This project investigates the differences between base and instruction-tuned large language models in terms of hallucination behaviour, evaluating their factual reliability and response calibration. It aims to apply and compare evaluation metrics and calibration techniques to better understand and mitigate hallucinations in generative NLP systems.
- Comparing Frequency-Based Extractive Summarisation with LLM- Based Abstractive Summarisation
  
  This project compares frequency-based extractive summarisation with LLM-based abstractive summarisation using the CNN/DailyMail dataset. The methods are evaluated using ROUGE and BERTScore, along with qualitative analysis of coherence, fluency, and content coverage.
- SCC.451 Coursework
  
  Use pretrained deep neural networks on image datasets, extract features, visualise them, cluster them, and classify images.Analyse the Seattle weather dataset by preprocessing the data and applying at least two clustering
- Hex SysOps
  
  Familiarisation with the Hex cluster and shadowing of SysOps maintenance work, with a view to contributing to ongoing operational tasks.