To Örebro University

Employee compliance with Information Security Policies (ISPs) depends on communicating clear and comprehensible content. However, existing research has shown that many ISPs are of poor communicative quality. Large Language Models (LLMs) could enhance ISPs if fine-tuned on high-quality data, but to do such fine-tuning requires a conceptual model for classifying the data and evaluating the resulting text. Therefore, as a step in this direction, the aim of this article is to develop a conceptual model of ISPs using Speech Act Theory as a theoretical lens to enable assessments of the communicative quality of ISPs. We used conceptual modeling and document analysis to develop the modelbased on 600 ISP statements from ten British National Health Service ISPs. We used selected parts from the SEQUAL framework to evaluate the model. The evaluation pointed to potential areas for improving the model’s semantic, empirical, physical, and deontic qualities. By incorporating these improvements, the final class diagram contains 21 classes, six of which address ISP statement quality as speech acts.

Antimicrobial resistance (AMR) poses a significant global health threat, impacting clinical treatments, agriculture, and public health. Although mass spectrometry techniques like MALDI-TOF provide opportunity for rapid AMR detection, current state-of-the-art models, such as MSDeepAMR, are limited to single-label classification, requiring separate models for each bacterium-antibiotic combination. These approaches struggle with challenges such as class imbalance, incomplete labels, and poor generalization across bacterial strains and antibiotics. This study addresses these limitations by introducing a novel multi-label, multi-bacteria classification framework (MLMBC) that simultaneously predicts AMR across multiple bacterial species and antibiotics using MALDI-TOF mass spectrometry data and Convolutional Neural Networks (CNNs), hereby establishing a solid foundation for the development of robust and rapid diagnostic tools to address the growing threat of multidrug-resistant bacteria. Utilizing CNNs in combination with transfer learning and semi-supervised classification techniques, such as self-training, pseudo-labeling and kNN label propagation, our approach improves accuracy and overcomes dataset limitations. Results demonstrate that the proposed approach with pseudo-labeling consistently outperforms single- and multi-label baselines; for example, it achieves AUROC >= 0.94 for E. coli, K.pneumoniae and S.aureus - Ceftriaxone pairs. Statistical tests support the effectiveness of the proposed approach, demonstrating that simultaneous multi label, multi-bacteria modeling constitutes a reliable and scalable strategy for antibiotic resistance prediction in clinically relevant settings.

Employees need clear guidance in order to contribute to their organ-isation’s digital sovereignty. Information security policies (ISPs) should provide employees with this guidance, but existing research show that this is not always the case. Large language models (LLMs) could enhance ISPs if finetuned on high-quality data, but such fine-tuning requires a robust way to distinguish between statements in ISPs. Therefore, and as a first step in this direction, the aim of this paper is to investigate the consistency when a Generative pre-trained transformer-like model classifies information security policy (ISP) statements into speech acts using a zero-shot learning (ZSL) approach. We investigate the consistency of LLAMA3.3-70B when applying ZSL to classify 600 selected ISP statements into speech act classes derived from Speech Act Theory. The findings show that LLAMA3.3-70B achieves high similarity to human classification and demonstrates strong consistency across multiple runs per different temperature settings. Furthermore, changes in the temper-ature setting had a minimal effect, which means that the model produces stable clas-sifications. We conclude that LLAMA3.3-70B is a useful starting point for continued work on improving ISP content using LLMs. 

To combat firms exploiting transfer pricing as a tool for tax avoidance, the Organisation for Economic Co-operation and Development (OECD) introduced the Base Erosion and Profit Shifting Action 13 (BEPS 13), enhancing tax transparency and the exchange of information with tax authorities. Interviewing senior-level transfer pricing advisors and tax officials, we investigate how and why multinational enterprises (MNEs) respond to this new regulation. By conducting this research, we contribute to the scarce literature assessing the impact of tax disclosure to tax authorities and, specifically, the implications of BEPS 13. The study finds that MNEs attach more importance to tax compliance and are more averse towards tax avoidance, which is mainly driven by higher audit pressure. Additionally, MNEs experience additional costs to comply with the different implementations in countries, centralise tax data with IT systems within the organisation, and set up multilateral Advance Pricing Arrangements (APAs) and Mutual Agreement Procedures (MAPs). To avoid (other forms of) tax avoidance being considered by companies, policymakers should address these concerns of increased compliance costs.

Employees compliance with information security policies (ISPs) depends on communicating clear and comprehensible content. However, existing research has shown that many ISPs are of poor communicative quality. Large language models (LLMs) could enhance ISPs if finetuned on high-quality data, but to do such fine-tuning requires a conceptual model for classifying the data and evaluating the resulting text. Therefore, as a step in this direction, the aim of this paper is to develop a conceptual model of ISPs using speech act theory as a theoretical lens. We use conceptual modelling and document analysis to develop the model and use selected parts from the SEQUAL framework to evaluate the model. Analysing 600 ISP statements from ten British National Health Service ISPs, we present a class diagram containing 19 classes, six of which address ISP statement quality as speech acts. The SEQUAL evaluation points to potential areas for improving the model’s semantic, empirical, physical and deontic qualities before using it to fine-tune LLMs to improve ISP content.