Modern slavery in global supply chains remains a critical concern, prompting governments in countries such as the UK, Australia, and Canada to introduce corporate transparency legislation, including Modern Slavery Acts. However, the growing volume of compliance reports has made manual analysis increasingly impractical. This challenge is compounded by the fact that many government agencies and NGOs responsible for assessing these statements operate in resource-constrained environments, limiting their ability to deploy large-scale language models. To address both issues, we present a novel AI framework for cross-jurisdictional compliance analysis. Our approach integrates four specialised teacher models, each trained using contrastive learning, prompt engineering, pseudo-labelling, and context-enhanced learning, which are distilled into a single lightweight, computationally efficient student model. This design enhances generalisability across legal frameworks while remaining affordable to deploy in low-resource settings. Experimental evaluations demonstrate that our method achieves higher accuracy and efficiency than existing approaches, offering a scalable, accessible solution for regulators and other stakeholders.

Named Entity Recognition (NER) from open-source security reports has become a crucial task in Cyber Threat Intelligence (CTI) to enable knowledge extraction and support proactive cyber defence. However, existing fully supervised NER approaches struggle to adapt to the dynamic and linguistically complex nature of CTI text. To address these challenges, we propose Label-Efficient Named Entity Recognition (LETNER), a model designed to handle multi-token, sparsely distributed, and fine-grained CTI entity patterns while maintaining low annotation demand. LETNER leverages Convolutional Neural Networks (CNNs) and a gating mechanism to learn dynamic span-based representations, and introduces an orthogonal regularisation to align entity-span relationships in a shared metric space for effective similarity-based inference. Furthermore, a cost-aware evaluation framework is presented to jointly quantify annotation effort and model performance, providing practical insights for decision-making in low-resource settings. Experimental results on a complex CTI dataset containing 22 fine-grained entity classes show that LETNER significantly outperforms baseline models, achieving high performance using only 10% of the annotated data.

Semi-supervised text classification has garnered significant attention for its ability to leverage unlabelled data in settings with limited labelled data. While current state-of-the-art methods employ consistency regularisation with pseudo-labels and augmentation techniques such as synonym replacement and back-translation, they often suffer from high computational cost and limited augmentation diversity. We propose SemiCR-VAE, a novel framework that uses a variational autoencoder with uncertainty-based sampling to generate diverse, task-aware augmentations from the latent space. Experiments on five datasets, including the newly introduced Hydrogen Energy dataset, show that SemiCR-VAE achieves the highest average accuracy (74%) and F1 score (76%), outperforming strong baselines by an average of 3% in accuracy and 4% in F1 score. Our approach exhibits robustness with tuned hyperparameters ( λ s = 100 , λ r l = 0.001 , λ k l = 10 , λ u = 0.5 , T = 0.7 ), ensuring effective text classification in low-resource settings with limited labelled data.

Multimodal learning aims to combine multiple modalities into a joint representation space. Existing works primarily focus on designing multimodal architectures to capture task-specific multimodal features. However, these methods often overlook variations in informativeness across modalities, leading to the feature-level bias problem, where irrelevant information introduces noise into the joint representation space. To address this issue, we propose a novel Deep Kernel Learning enhanced Multimodal Classification (DKLMC) framework. The framework comprises a main network for multimodal classification and a deep kernel learning network for estimating the unimodal contribution for each instance. The DKLMC framework is capable of estimating the contribution of each modality for each instance based on its unimodal contribution factor. This factor is derived using an auxiliary network, and each modality's informativeness is quantified by incorporating the uncertainty of the prediction confidence. Specifically, it addresses the feature-level bias problem by emphasising the more informative unimodal feature and reducing the impact of less informative features in the joint representation space. It features two key innovations: (1) an auxiliary network based on deep kernel learning to estimate unimodal contribution factor effectively, and (2) the use of the Reptile algorithm, a gradient-based meta-learning approach that aims to improve adaptability across various tasks, to balance the optimisation between main and auxiliary networks, ensuring effective learning of both unimodal contribution estimation and multimodal classification tasks. Extensive experiments across diverse datasets demonstrate the effectiveness of the proposed DKLMC framework.

Modern Slavery Acts mandate that corporations disclose their efforts to combat modern slavery, aiming to enhance transparency and strengthen practices for its eradication. However, verifying these statements remains challenging due to their complex, diversified language and the sheer number of statements that must be reviewed. The development of NLP tools to assist in this task is also difficult due to a scarcity of annotated data. Furthermore, as modern slavery transparency legislation has been introduced in several countries, the generalizability of such tools across legal jurisdictions must be studied. To address these challenges, we work with domain experts to make two key contributions. First, we present AIMS.uk and AIMS.ca, newly annotated datasets from the UK and Canada to enable cross-jurisdictional evaluation. Second, we introduce AIMSCheck, an end-to-end framework for compliance validation. AIMSCheck decomposes the compliance assessment task into three levels, enhancing interpretability and practical applicability. Our experiments show that models trained on an Australian dataset generalize well across UK and Canadian jurisdictions, demonstrating the potential for broader application in compliance monitoring. We release the benchmark datasets and AIMSCheck to the public to advance AI-adoption in compliance assessment and drive further research in this field.

Multimodal learning for classification tasks has recently gained significant attention in bioinformatics. Current approaches primarily concentrate on devising efficient deep learning architectures to capture features within and across modalities. However, they typically assume that each modality contributes equally to the classification objective, overlooking inherent biases within multimodal learning. This paper presents a modality contribution confidence-enhanced deep learning framework to address this issue, resulting in an improved fusion space and improved classification performance on multiomics data. Specifically, we propose utilising a non-parametric Gaussian Process to assess the unimodal confidence of each modality and learn within-modality features. Additionally, we introduce the use of the Kullback-Leibler divergence to align multiple modalities and learn cross-modality features. Extensive experiments on four multiomics datasets, incorporating modalities such as static information, DNA, mRNA, miRNA, and protein data, validate the effectiveness of the proposed method. Furthermore, a case study on the blister recovery task is included to demonstrate the practical utility of our model.

Abstract The world of data is vast and complex, harbouring valuable insights and patterns that can drive decision-making in various fields. However, accessing useful information from raw data (e.g. mining project reports) can be a formidable task, often requiring specialised skills and tools. The emergence of generative artificial intelligence (AI) has opened up an intriguing and novel means of engaging with data conversation. This article delves into the novel concept of chatting with organisational data using generative AI. We present an innovative solution that combines a generative AI chatbot (e.g. ChatGPT-QAM) with a language model (e.g. BERT) based extractive question-answer model (BERT-QAM) to generate responses based on given contexts. We use an answer verification model to resolve any disagreements between the responses of ChatGPT-QAM and BERT-QAM. We use the context filtering model to enhance responses by considering valid contexts. Our solution is tested on mining project reports made available by the Geological Survey of Queensland. Through this case study, we highlight several challenges that must be addressed to utilise this approach effectively. The case study shows that the concept of chatting with organisational data can revolutionise how we interact with complex scientific reports, which contain a mix of tables, text and images, to find valuable insights.

• This paper proposes Latent Space Refinement (LSR), a novel unsupervised classification framework that integrates metric learning with clustering-based representation refinement, addressing the critical challenge of label scarcity in cyber threat intelligence (CTI). • LSR introduces a posterior regularisation strategy that aligns latent representations from Pretrained Language Models (PLMs) with an auxiliary TF-IDF-based distribution. This guides unsupervised adaptation to the target domain without any PLM fine-tuning, ensuring scalability and efficiency. • Extensive experiments on three CTI benchmarks demonstrate that LSR consistently outperforms state-of-the-art unsupervised and few-shot baselines in Accuracy and F1 score. • By enabling lightweight unsupervised domain adaptation, LSR offers a plug-and-play solution applicable to CTI, including other resource-constrained domains such as health and legal text classification. Text classification plays a critical role in Cyber Threat Intelligence (CTI) applications, where open-source text data is mined to identify patterns such as Indicators of Compromise (IoC), Tactics, Techniques and Procedures (TTPs), Named Entities and more. However, the dynamic nature of CTI makes traditional supervised machine learning classifiers impractical due to their reliance on large number of labelled training datasets. To address this, we propose Latent Space Refinement (LSR), an unsupervised method designed for CTI text classification. LSR introduces a posterior regularisation strategy where an auxiliary distribution derived from a TF-IDF feature space serves as signals to refine latent representations derrived from Pretrained Language Models (PLMs). By iteratively refining this latent space with clustering signals, LSR enables efficient similarity-based classification using only a few user-provided seed keywords. Extensive experiments on diverse CTI tasks, including both binary and multi-class classification, demonstrate that LSR consistently outperforms state-of-the-art unsupervised and zero-shot/few-shot methods in Accuracy and Weighted F1 score, all without tuning internal PLM parameters. This makes LSR a lightweight and PLM-agnostic solution for real-world CTI applications.

Multi-modal data fusion for effective feature representation in machine learning is challenging due to intrinsic biases present within and across different modalities. Existing multi-modal data fusion methods often face difficulties in learning generic features due to diverse noise patterns and variations in feature dynamics across different modalities. In this paper, we present a novel method called Uncertainty-guided Meta-Learning Multi-modal Fusion and Classification (UMLMC) to address these challenges. UMLMC dynamically transforms multi-modal feature spaces at both the pre- and post-fusion levels by incorporating uncertainty estimates from an auxiliary network. Our model is optimized using a meta-learning algorithm to enhance its generalization capabilities. Extensive experiments on multi-modal data from diverse domains, along with comparisons to state-of-the-art methods, demonstrate the effectiveness of UMLMC in improving classification performance. These results confirm that UMLMC, with its innovative uncertainty estimation and meta-learning framework, effectively learns informative intra- and inter-modal features, leading to superior classification outcomes. • An uncertainty-guided meta-fusion method for multi-modal fusion and classification. • Mitigating the impact of feature-level bias at both before and after fusion. • Meta-learning to generate less biased uncertainty estimation at the feature level.

Abstract Generative adversarial networks (GANs) have demonstrated their effectiveness in generating temporal data to fill in missing values, enhancing the classification performance of time series data. Longitudinal datasets encompass multivariate time series data with additional static features that contribute to sample variability over time. These datasets often encounter missing values due to factors such as irregular sampling. However, existing GAN-based imputation methods that address this type of data missingness often overlook the impact of static features on temporal observations and classification outcomes. This paper presents a novel method, fusion-aided imputer-classifier GAN (FaIC-GAN), tailored for longitudinal data classification. FaIC-GAN simultaneously leverages partially observed temporal data and static features to enhance imputation and classification learning. We present four multimodal fusion strategies that effectively extract correlated information from both static and temporal modalities. Our extensive experiments reveal that FaIC-GAN successfully exploits partially observed temporal data and static features, resulting in improved classification accuracy compared to unimodal models. Our post-additive and attention-based multimodal fusion approaches within the FaIC-GAN model consistently rank among the top three methods for classification.

Multi-modal representation learning has received significant attention across diverse research domains due to its ability to model a scenario comprehensively. Learning the cross-modal interactions is essential to combining multi-modal data into a joint representation. However, conventional cross-attention mechanisms can produce noisy and non-meaningful values in the absence of useful cross-modal interactions among input features, thereby introducing uncertainty into the feature representation. These factors have the potential to degrade the performance of downstream tasks. This paper introduces a novel Pre-gating and Contextual Attention Gate (PCAG) module for multi-modal learning comprising two gating mechanisms that operate at distinct information processing levels within the deep learning model. The first gate filters out interactions that lack informativeness for the downstream task, while the second gate reduces the uncertainty introduced by the cross-attention module. Experimental results on eight multi-modal classification tasks spanning various domains show that the multi-modal fusion model with PCAG outperforms state-of-the-art multi-modal fusion models. Additionally, we elucidate how PCAG effectively processes cross-modality interactions.

Early classification of longitudinal data remains an active area of research today. The complexity of these datasets and the high rates of missing data caused by irregular sampling present data-level challenges for the Early Longitudinal Data Classification (ELDC) problem. Coupled with the algorithmic challenge of optimising the opposing objectives of early classification (i.e., earliness and accuracy), ELDC becomes a non-trivial task. Inspired by the generative power and utility of the Generative Adversarial Network (GAN), we propose a novel context-conditional, longitudinal early classifier GAN (LEC-GAN). This model utilises informative missingness, static features and earlier observations to improve the ELDC objective. It achieves this by incorporating ELDC as an auxiliary task within an imputation optimization process. Our experiments on several datasets demonstrate that LEC-GAN outperforms all relevant baselines in terms of F1 scores while increasing the earliness of prediction.

Language model (LM) has become a common method of transfer learning in Natural Language Processing (NLP) tasks when working with small labeled datasets. An LM is pretrained using an easily available large unlabelled text corpus and is fine-tuned with the labelled data to apply to the target (i.e., downstream) task. As an LM is designed to capture the linguistic aspects of semantics, it can be biased to linguistic features. We argue that exposing an LM model during fine-tuning to instances that capture diverse semantic aspects (e.g., topical, linguistic, semantic relations) present in the dataset will improve its performance on the underlying task. We propose a Mixed Aspect Sampling (MAS) framework to sample instances that capture different semantic aspects of the dataset and use the ensemble classifier to improve the classification performance. Experimental results show that MAS performs better than random sampling as well as the state-of-the-art active learning models to abuse detection tasks where it is hard to collect the labelled data for building an accurate classifier.

A new technique is introduced to model misperception by participating decision makers (DMs) in a conflict having two or more DMs within the framework of the graph model for conflict resolution. This comprehensive approach enables one to model a conflict situation involving misperception: held by and about the focal DM and its opponents. To achieve this, DMs' options in a conflict situation are classified based on different kinds of misperception that can alter the choices of the focal DM and/or the other DMs. Furthermore, the combination of DMs' options can generate the universal set of options for the entire conflict, which can then be used to construct the universal set of states. This novel design can differentiate between the states that are recognized by all DMs and those that are recognized individually. Furthermore, eight sets of equilibria are formally defined within the construction of the first-level hypergame in graph form to provide strategic insights into the conflict and reflect the effect of DMs' misperceptions on the equilibria of the dispute.

Abstract Patterns are used as a fundamental means for analyzing data in many data mining applications. Many efficient techniques have been developed to discover patterns. However, the excessive number of discovered patterns and the lack of semantic information have made it difficult for a user to interpret and explore the patterns. A rough idea of the meanings of patterns can benefit the user in the process of exploring them. To address this issue, this paper presents a model for automatically annotating patterns with concepts. In addition, in a given context, the relative importance of each term that defines a concept is not the same. To define a context, there are a number of related information sources, such as documents, patterns, concepts, and an ontology. The question is which information sources are useful for estimating the relative importance of the terms? Should the most accurate one to be focused on or all of them be used to define the context? This research investigated these questions and defined an effective annotation context to estimate the relative importance of the terms, where the aim is to improve the performance of a machine that relies on the subject matter of a pattern set. The model is evaluated by comparing it with different baseline models on 2 standard datasets. The results show that the performance of the proposed model is significantly better.

Topic modelling methods such as Latent Dirichlet Allocation (LDA) have been successfully applied to various fields, since these methods can effectively characterize document collections by using a mixture of semantically rich topics. So far, many models have been proposed. However, the existing models typically outperform on full analysis on the whole collection to find all topics but difficult to capture coherent and specifically meaningful topic representations. Furthermore, it is very challenging to incorporate user preferences into existing topic modelling methods to extract relevant topics. To address these problems, we develop a novel personalized Association-based Topic Selection (ATS) model, which can identify semantically valid and relevant topics from a set of raw topics based on the semantical relatedness between users’ preferences and the structured patterns captured in topics. The advantage of the proposed ATS model is that it enables an interactive topic modelling process driven by users’ specific interests. Based on three benchmark datasets, namely, RCV1, R8, and WT10G under the context of information filtering (IF) and information retrieval (IR), our rigorous experiments show that the proposed ATS model can effectively identify relevant topics with respect to users’ specific interests, and hence to improve the performance of IF and IR.

The 1956 nationalization of the Suez Canal dispute, between Egypt and an alliance between the United States (US) and Britain, is modeled and analyzed within the paradigm of a first-level hypergame in graph form. The results reveal that Britain and US encountered a strategic surprise in the dispute on account of their failure to predict Egypt's action in nationalizing the Suez Canal. The equilibrium states of the dispute are analyzed to provide insights into the types of the misperceptions that caused the conflict, the possible decision makers' reactions after observing the equilibrium states, and the sustainability of the first-level hypergame equilibria. The analysis reveals substantial motivation for the United States and Britain to intensify the dispute to recapture the Suez Canal and to improve their situation.

Understanding or acquiring a user's information needs from their local information repository (e.g. a set of example-documents that are relevant to user information needs) is important in many applications. However, acquiring the user's information needs from the local information repository is very challenging. Personalised ontology is emerging as a powerful tool to acquire the information needs of users. However, its manual or semi-automatic construction is expensive and time-consuming. To address this problem, this paper proposes a model to automatically learn personalised ontology by labelling topic models with concepts, where the topic models are discovered from a user's local information repository. The proposed model is evaluated by comparing against ten baseline models on the standard dataset RCV1 and a large ontology LCSH. The results show that the model is effective and its performance is significantly improved.

A methodology is developed to model a decision maker's (DM's) fuzzy and/or interval fuzzy preference over feasible scenarios or states within the framework of the graph model for conflict resolution. This technique uses the DM's fuzzy relative importance of its preference statements and their fuzzy truth values for the feasible states in a conflict under uncertain conditions. A preference statement of a DM is a preferable combination of DM's options or courses of action. The fuzzy importance for one preference statement over another, a value in the interval [0, 1], is interpreted as the degree to which the first preference statement is more important than the second to the DM. A fuzzy truth value of a preference statement at a feasible state is a number in the interval [0, 1] that represents the degree to which the statement is true at the state. When the DM is confident in its pairwise fuzzy importance degrees over the preference statements and their fuzzy truth values at the feasible states, the methodology provides a fuzzy preference over the states. When there is an ordinary or crisp importance ordering of preference statements and when the truth values of preference statements are classical or crisp at the feasible states, the technique generates a crisp preference over the states. The methodology is illustrated using a case study.

To capture uncertainty in preferences, definitions based on grey numbers are incorporated into the graph model for conflict resolution (GMCR), a realistic and flexible methodology to model and analyze strategic conflicts. A general grey number is a real number that may be a member of a discrete set of real numbers, or may fall within one or several intervals. It can represent uncertain preference of decision makers in a meaningful way. Here, a grey-based preference structure is developed and integrated with GMCR. Utilizing a number of grey-based ideas, solution concepts describing human behavior under conflict in the face of uncertain preference are defined for a conflict model. This grey-based GMCR is then applied to a generic sustainable development conflict with uncertain preferences in order to demonstrate how it can be conveniently utilized in practice.

A generic sustainable development conflict having uncertain preferences is investigated using a new methodology. In particular, definitions of grey numbers are introduced for incorporation into the Graph Model for Conflict Resolution (GMCR) in order to capture uncertainty in preferences. The GMCR constitutes a simple and flexible methodology model and analyze strategic conflicts. The goal of the study is to analyze the influence of uncertainty on the outcome of the conflict and to provide suggestion on guiding the conflict to move in a desired direction. General grey numbers are used to represent the uncertain preferences of decision makers. Following the explanation of four solution concepts based on grey preferences, a stability analysis is conducted and some new insights are provided in this research.

A methodology is developed to use the fuzzy truth values of preference statements for feasible states in an option prioritization technique in order to rank states within the framework of the Graph Model for Conflict Resolution. Option prioritization ranks states based on the truth values of preference statements, which are compositions of the decision makers' courses of actions joined by logical connectives, ordered lexicographically. Fuzzy truth values are represented as truth degrees; so they include binary truth values, “true” and “false”, as well as other possible truth intensities that are reasonable according to the specific circumstances. Consequently, the assumption of fuzzy truth values of preference statements provides more realistic preference ordering of feasible states. The methodology is applied to the Elmira groundwater contamination dispute, which took place in Elmira, Ontario, Canada, for eliciting the preferences of decision makers, to demonstrate the applicability of this technique.

A new framework for the graph model for conflict resolution is developed so that decision makers (DMs) with fuzzy preferences can be included in conflict models. A graph model is both a formal representation for multiple participant-multiple objective decision problems and a set of analysis procedures that add insights into them. Within the new framework, graph models can include-and integrate into the analysis-both certain and uncertain information about DMs' preferences. One key contribution of this study is to extend the four basic stability definitions for two or more DMs to models with fuzzy preferences. Together, fuzzy Nash stability, fuzzy general metarationality, fuzzy symmetric metarationality, and fuzzy sequential stability provide anuanced description of human behavior. A state is fuzzy stable for a DM if a move to any other state is not sufficiently likely to yield an outcome which the DM prefers, where sufficiency is measured according to a fuzzy satisficing threshold that is the characteristic of the DM. A fuzzy equilibrium, which is an outcome that is fuzzy stable for all DMs, therefore represents a possible resolution of the strategic conflict. The practical application and interpretation of these new stability definitions are illustrated with an example.

A Fuzzy Preference Framework for the Graph Model for Conflict Resolution is employed to analyze the sustainable development conflict in order to gain strategic insights. This framework is a novel methodology to identify outcomes favorable for all parties in a multiple-participant multiple-objective decision making setting with both certain and uncertain preference information. When applied to the sustainable development conflict, in which a developer plans to build or operate a project inspected by an environmental agency, the methodology identifies stable outcomes and thus clarifies the necessary conditions for sustainability.

A fuzzy preference framework is developed within the paradigm of the graph model for conflict resolution. This framework takes into account both certain and uncertain information about the preferences of decision makers (DMs) involved in a strategic conflict. The graph model is a solution methodology for conflict decision making that begins with a model of the problem and suggests possible resolutions through a number of stability definitions. Four basic fuzzy stability definitions are introduced for a two-DM graph model to analyze conflict behavior and identify possible resolutions even when preferences are fuzzy. Fuzzy stability definitions describe varied human behavior in a conflict model; a state is fuzzy stable for a DM according to a specific fuzzy stability definition if a move to any other state, evaluated according to that definition, does not meet the DM's fuzzy satisficing threshold. A state that is fuzzy stable for all DMs under a specific fuzzy stability definition constitutes a fuzzy equilibrium under that definition, and is interpreted as a possible resolution of the conflict. Fuzzy stability definitions include fuzzy Nash stability, fuzzy general metarationality, fuzzy symmetric metarationality, and fuzzy sequential stability.

A systematic study of recent developments in preferences based on fuzzy logic is carried out in order to identify an effective design for fuzzy preference models in conflict resolution. Fuzzy preference, defined via a fuzzy relation over the alternatives or states, attempts to represent a decision maker's preferences more realistically. It generalizes the usual preference structures in the sense that it provides a uniform description of both certain (crisp) and uncertain preferences. The potential applicability of fuzzy preferences to the modeling, analysis, and understanding of strategic conflicts is investigated and connected to a literature survey.