BURA Collection:

Ontology-based Keyword Enriched Topic Modelling for Biomedical Research Trends: A Business Intelligence Perspective

2026-06-28T02:00:43Z

Title: Ontology-based Keyword Enriched Topic Modelling for Biomedical Research Trends: A Business Intelligence Perspective Authors: Altarawneh, A; Tucker, A; Swift, S; Arzoky, M Abstract: Over the past ten years, the biomedical literature has increased exponentially, and it is becoming increasingly cumbersome for the researchers and policymakers to identify emerging directions in research and allocate resources appropriately. Bibliometric indicators that are traditional and topic models based on keywords have weaknesses of synonym ambiguity, lag times and shallow semantics. The paper presents a new pipeline, which adds hierarchical concepts of the Medical Subject Headings (MeSH) ontology to author keywords and then uses Latent Dirichlet Allocation (LDA) to identify topics. A data set of 624,555 journal articles from 2015 to 2024 has been collected from the Scopus API in medicine, health professions and molecular biology. Four different systematic representations, raw keywords, MeSH parent categories, direct MeSH matches, and fully enriched keywords are pre-processed. The results showed that using fully enriched keywords with MeSH ontology annotations significantly enhanced topic coherence from 0.338 (baseline keywords) to 0.574, producing more interpretable and semantically consistent topics. The spatial analysis demonstrates another evidence of changing research priorities at the geographical level. The suggested framework illustrates the strength of ontology-driven modelling and visual analytics to assist in making business-intelligence decisions in rapidly evolving biomedical settings.

Ontology-Enriched Abstract Topic Modelling for Biomedical Trend Discovery: A MeSH-based LDA Framework with BI Visual Analytics

2026-06-28T02:00:42Z

Title: Ontology-Enriched Abstract Topic Modelling for Biomedical Trend Discovery: A MeSH-based LDA Framework with BI Visual Analytics Authors: Altarawneh, A; Swift, S; Tucker, A; Arzoky, M Abstract: The explosive growth of biomedical literature increasingly makes it difficult to identify incipient topics and long-term developments. Citation-based indicators are slow to capture developments and often fail to represent influence. Whilst Latent Dirichlet Allocation (LDA) supports content-based trend discovery, its bag-of-words assumption can overlook synonymy, polysemy, and hierarchical relations that are central to biomedical terminology. Moreover, recent work provides limited direct evidence on how far MeSH-enriched abstract representations improve LDA-based trend analysis when compared against the original abstract text. To fill this gap, an ontology-driven methodology for the pre-topic modelling of article abstracts by enriching them with hierarchical Medical Subject Headings (MeSH) terms is introduced in this study. We evaluate four representations, namely original abstracts, MeSH parent categories, MeSH-matched terms with parents, and a fully enriched representation, using a decade-long Scopus corpus of 624,486 biomedical journal articles (2015–2024). Model performance is assessed using coherence, distinctiveness, perplexity, and stability. The fully enriched representation achieves the strongest overall results (coherence 0.606, distinctiveness 0.996, stability 0.725, perplexity 1782.040) relative to the original abstract baseline. Temporal topic analysis further reveals interpretable patterns, including the COVID-19 surge and longer-term structural shifts such as the rise of machine-learning diagnostics. These findings indicate that MeSH-guided abstract enrichment can strengthen topic quality and support more effective biomedical trend discovery through business-intelligence visual analytics.

The Goofy Game: an Approach to Medical AI Misalignment

2026-06-27T08:59:57Z

Title: The Goofy Game: an Approach to Medical AI Misalignment Authors: Puccio, B; Castagna, F; Tucker, A; Veltri, P Abstract: While Large Language Models (LLMs) offer transformative potential across domains, often outperforming human benchmarks in various tasks, they remain vulnerable to exploitation by users aiming to override their safety protocols. Despite the progress achieved through red teaming methodologies in uncovering and mitigating such vulnerabilities, one notably persistent technique, referred to here as the “Goofy Game”, which leverages role-playing strategies, continues to bypass many existing safeguards. This technique can elicit unsafe responses from LLMs, which, although seemingly benign in isolation, could lead to severe consequences when deployed within high-stakes environments such as clinical decision-making or patient communication. In this study, we build on the insights from our previous exploratory experiments and analyse how a malicious user, even without technical knowledge of the internal architecture and parameters of generative AI models, could create a role-playing prompt that coerces a language model (LLM) into generating incorrect and potentially harmful clinical suggestions. Our objective is to elucidate a particular vulnerability scenario and provide insights that will contribute to future advancements in the development of secure and reliable AI systems.

Assessing the impact of synthetic data generated by Bayesian networks on heart disease prediction

2026-06-27T02:00:42Z

Title: Assessing the impact of synthetic data generated by Bayesian networks on heart disease prediction Authors: Lazzaro, I; Milano, M; Tucker, A; Cannataro, M Abstract: Synthetic data generation using Bayesian networks (BN) offers a promising approach to overcoming data scarcity in clinical prediction tasks, yet its actual impact on model performance remains underexplored. This study investigates the use of Bayesian network-based generative models to produce synthetic patient data and examines how the quality of the original real data influences the effectiveness of such augmentation. Three benchmark datasets from the UCI Heart Disease repository (Cleveland, Hungary, and Switzerland) were employed, all sharing an identical structure comprising 13 clinical predictors. The Cleveland dataset, which is the most complete and consistent among the three, was used exclusively as the training source for learning the Bayesian network structure and parameters under clinically informed constraints. To ensure robust evaluation, the dataset was partitioned into two independent subsets: 153 patients were used to train the Bayesian network, while 150 held-out patients were used exclusively to generate synthetic records. Predictive models were trained under three configurations: real data only, synthetic data only, and a hybrid real + synthetic (filtered) dataset, and evaluated using 10-fold cross-validation and external validation on independent cohorts. Results indicate that integrating real and synthetic data significantly improved accuracy and precision, particularly for the Switzerland cohort (F(2,27)=23.06, η²=0.63)), whereas improvements were smaller and partially non-significant in the noisier Hungarian dataset. These findings demonstrate that the effectiveness of synthetic augmentation depends on the structure and completeness of the source data, underscoring the importance of data quality for reliable generative modelling in clinical prediction. Description: Data availability: All datasets used in this work are freely available in the UCI repository.