Domain Knowledge-Enhanced LLMs for Fraud and Concept Drift Detection

Phishing attacks continue to grow in scale and sophistication, causing substantial financial losses and privacy breaches worldwide. Recent advances in large language models (LLMs) have brought significant changes to the generation and detection of phishing content. This study systematically investigates the dual role of LLMs in facilitating phishing attacks and strengthening countermeasures. Using the PRISMA methodology, authors screened 142 records published between January 2023 and April 2025 and identified 36 eligible studies from major academic databases, including IEEE Xplore, ScienceDirect, ACM Digital Library, Web of Science, and Scopus. A comprehensive and rigorous analysis was conducted of research trends/themes over time, dataset characteristics, and the LLM architectures/models employed. The findings reveal that most studies relied on manually generated datasets rather than publicly available benchmark datasets, and that GPT-based models received considerably more attention than other LLM architectures. The review demonstrates that LLMs substantially enhance the generation of phishing content by producing coherent, contextually relevant, and persuasive email and website content. This capability lowers the technical barrier for attackers and potentially increases attack effectiveness. Conversely, LLMs also strengthen defensive strategies by enabling more effective analysis of textual and visual content for phishing detection. In many cases, LLM-based approaches outperform traditional machine learning and deep learning methods and, in certain contexts, approach or match human-level performance. Overall, the findings suggest that LLMs have accelerated and automated phishing-related processes, simultaneously intensifying the threat landscape and advancing defensive capabilities. • The first in-depth study to review LLMs usage in Phishing attack generation and detection. • The study reveals LLMs have accelerated and automated phishing-related processes, elevating both threats and defence mechanisms. • GenAI-based multimodal phishing attacks are on the rise due to the wider adoption of GenAI tools in general.

Model tuning or prompt Tuning? a study of large language models for clinical concept and relation extraction

The rapid advancement of artificial intelligence (AI) technologies, particularly generative AI and large language models (LLMs), has ushered in a new era of possibilities for the financial sector. This paper explores the integration of these cutting-edge technologies into financial sector risk management, examining both the potential applications and the necessary regulatory frameworks. We provide a comprehensive analysis of how generative AI and LLMs can revolutionize risk assessment, fraud detection, market analysis, and regulatory compliance. The study delves into the technical aspects of these AI models, their implementation challenges, and the implications for existing risk management practices. Furthermore, we propose a novel framework for the responsible adoption of AI in financial risk management, addressing concerns related to model interpretability, data privacy, and algorithmic bias. Our findings suggest that while generative AI and LLMs offer unprecedented opportunities for enhancing risk management capabilities, they also necessitate a recalibration of regulatory approaches to ensure financial stability and consumer protection. This research contributes to the growing body of literature on AI in finance and provides actionable insights for practitioners, policymakers, and researchers in the field.

Record linkage is an essential task in data integration in the fields of healthcare, law enforcement, fraud detection, transportation, biology, and supply chain management. The problem of record linkage is to cluster records from various sources such that each cluster belongs to a single entity. Scalability in record linking is limited by the large number of pairwise comparisons required. Blocking addresses this challenge by partitioning data into smaller parts, substantially reducing the computational cost. With the advancement of Large Language Models (LLMs), there are several possibilities to improve record linkage by leveraging their semantic understanding of textual attributes. LLM-based record linkage algorithms in the literature have very large runtimes. In this paper, we show that the employment of blocking can result in significant improvements not only in the runtime but also in the accuracy. Specifically, we propose a record linkage algorithm that combines LLMs with blocking. Experimental evaluation demonstrates that our algorithm achieves lower runtimes while simultaneously improving F1 scores compared to the approaches relying solely on LLMs. These findings demonstrate the importance of blocking even in the era of advanced machine learning models.

Large Language Models (LLMs) excel in fields such as natural language understanding, generation, complex reasoning, and biomedicine. With advancements in materials science, traditional manual annotation methods for phase diagrams have become inadequate due to their time-consuming nature and limitations in updating thermodynamic databases. To overcome these challenges, we propose a framework based on instruction tuning, utilizing LLMs for automated end-to-end annotation of phase diagrams. High-quality phase diagram images and expert descriptions are collected from handbooks and then preprocessed to correct errors, remove redundancies, and enhance information. These preprocessed data form a golden dataset, from which a subset are used to train LLMs through hierarchical sampling. The fine-tuned LLM is then tested for automated phase diagram annotation. Results show that the fine-tuned model achieves a cosine similarity of 0.8737, improving phase diagram comprehension accuracy by 7% compared to untuned LLMs. To the best of our knowledge, this is the first paper to propose using LLMs for the automated annotation of phase diagrams, replacing traditional manual annotation methods and significantly enhancing efficiency and accuracy.

The purpose of this study was to empirically examine and accurately synthesise the effect of auditor quality, internal auditor professional ethics and internal auditor independence on the fraud detection process with information technology as a moderating variable. The process of detecting financial and economic fraud in a company plays an important role in the internal control system. The population of this study consists of internal auditors from companies located in Jakarta and Bogor. The results of this study were conducted using reliability and validity tests, with data analysis performed through descriptive analysis with three dependent variables: the Quality of Auditors as the first dependent variable (X1), the Professional Ethics as the second dependent variable (X2), and the Independence as the third dependent variable (X3). Fraud Detection serves as the independent variable (Y), while Information Technology acts as the moderating variable (Z). The results of this study indicate that the influence of auditor quality on fraud detection is positive and significant, the influence of professional ethics on fraud detection is positive and significant, the influence of independence on fraud detection is positive and significant, and the role of information technology in the influence of audit quality is positive but not significant, the role of information technology in the influence of professional ethics on fraud detection is positive but not significant, and the influence of independence on fraud detection through information technology as a moderating variable is positive and significan. This research can have implications for improving fraud prevention and detection by comprehensively improving audit quality, professional ethics and internal auditor independence by maximising the use of information technology Keywords: Auditor Quality, Profesional Ethics, Independenc and Information Technology

The massive digitalization of the communication environment has created fertile ground for increasingly sophisticated social engineering attacks, which exploit human psychology rather than technical flaws. Online criminal activity: Today, fraudsters hijack real-time conversations across voice calls, chat platforms, and emails to trick individuals and organizations into divulging private information or approving fraudulent transactions. Traditional fraud detection strategies, which rely mostly on static keyword-based heuristics or predefined rule-based detection, have proven to be less effective against these dynamic and adaptive threats. Therefore, in this paper, we propose a real-time fraud detection model based on fine-tuned large language models (LLMs) to bridge this gap. Unlike conventional systems, the proposed architecture leverages deep contextual understanding, semantic reasoning, and intent classification to identify suspicious interactions in live communication environments. The system integrates several key components: a speech-to-text transcription pipeline for converting voice calls into structured text; a retrieval-augmented generation (RAG) mechanism that incorporates organizational policies and domain-specific knowledge into decision-making; and a feedback loop enabling continuous adaptation to novel fraud strategies. In addition, the framework utilizes a scenario generator to augment the datasets, generating contrastive benign versus malicious dialogue as a means to enhance model robustness. Trained with LoRA and quantization techniques for efficiency reasons, the model performs well on controlled evaluations, reaching over 97% accuracy in predicting the intent of fraudulent messages within three conversational turns. Real-world deployment results demonstrate tangible improvements in incidents related to fraud, enhanced decision support for analysts utilizing outputs from explainable AI, and increased flexibility in responding to new threats. In addition to advancing the technical state of fraud detection research, this work also contributes to the broader research efforts on cybersecurity resiliency by demonstrating the feasibility of operationalizing LLMs for high-stakes real-time applications.

Knowledge base question answering (KBQA) refers to the system that produces answers to user queries by reasoning with a large-scale structured knowledge base. Advanced works have achieved great success either by generating logical forms (LF) or directly generating answers. Although the former typically yields better performance, these generated LF could be inaccurate, e.g., non-executable. In this regard, large language models (LLMs) have shown exciting potential for accurate generation. However, it is challenging to fine-tune LLMs to generate LF. This is because the context retrieved for prediction typically leads to an excessive number of reasoning paths. In this context, LLMs can generate numerous LF corresponding to these reasoning paths, but a few LF can result in correct answers. Thus, fine-tuning LLMs to generate answer-relevant LF would conflict with the prior knowledge of the LLMs. In this work, we propose a novel learning framework, FM-KBQA, to fine-tune LLMs using multi-task learning for KBQA. Specifically, we propose to fine-tune LLMs using an additional objective: generating the index of reasoning paths that lead to correct answers. This will direct LLMs to pay attention to answer-relevant paths among numerous reasoning paths by completing a simple task where the selected reasoning paths can be supplementary for non-executable LF. Directly generating answers can make LLMs pay attention to the answer-relevant reasoning paths, but it is much more challenging than generating the index of reasoning paths. To verify FM-KBQA's effectiveness, we conduct experiments on mainstream benchmarks, such as WebQuestionsSP (WQSP) and ComplexWebQuestions (CWQ). Extensive evaluations across two public benchmark datasets underscore the superiority of FM-KBQA over current state-of-the-art methods.

Purpose With accelerating advances in artificial intelligence (AI) and digital transformation, the construction domain is rapidly adopting large language models (LLMs) to improve efficiency, safety and sustainability. However, a comprehensive understanding of their applications, adaptation strategies and evaluation practices in this domain remains limited. This paper provides a systematic landscape of LLM research and applications in construction, clarifying their roles, integration pathways and associated challenges. Design/methodology/approach A systematic review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guideline, combining bibliometric and content analysis of publications from 2020 to 2025. The review examines the evolution, implementation strategies and performance characteristics of LLMs in construction domains. Findings LLMs have been increasingly implemented across diverse construction tasks. Most studies utilize pretrained models, adapted through fine-tuning, retrieval-augmented generation or prompt engineering. While LLMs demonstrate strong reasoning and semantic understanding capabilities, challenges persist in domain adaptation, hallucination control, data privacy and lifecycle integration. Practical implications The results provide insights for researchers and practitioners on how LLMs can be effectively implemented throughout the construction domain. The paper highlights the need for developing domain-specific evaluation standards, establishing governance frameworks and aligning LLM applications with existing digital ecosystems such as building information modeling and digital twins. Originality/value This paper presents a comprehensive systematic review focused on LLM applications in construction. By mapping the technological landscape and summarizing key implementation strategies, it establishes a theoretical and practical foundation for advancing intelligent, data-driven and human-centered construction in the era of generative AI.

As generative artificial intelligence (AI) tools become widely adopted, large language models (LLMs) are increasingly used on both sides of high-stakes decision processes, ranging from hiring to content moderation. This dual adoption raises a critical question: do LLMs systematically favor content that resembles their own outputs? Prior research in computer science has identified self-preference bias---the tendency of LLMs to favor their own generated content---but its real-world implications have not been empirically evaluated. We focus on the hiring context, where job applicants increasingly use LLMs to refine resumes, while employers integrate LLMs into their recruitment pipelines to screen those same resumes. Using a controlled resume correspondence experiment, we find that LLMs consistently prefer resumes generated by themselves over those written by humans or by alternative models, even when content quality is controlled. The bias against human-written resumes is particularly substantial, with self-preference rates ranging from 68 percent to 92 percent across a diverse set of commercial and open-source models. This behavior introduces a novel form of algorithmic unfairness, one that advantages users of specific AI tools and disadvantages others based on their tool choices or access. We further show that this bias can be significantly reduced through simple interventions based on LLMs' self-recognition capabilities, yielding reductions of over 60 percent. These findings highlight an emerging but previously overlooked risk in AI-assisted decision making and call for expanded frameworks of AI accountability that address not only demographic disparities but also biases in model-to-model interaction.

In evolving data stream, when its concept undergoes a change it is known as concept drift. Detecting concept drift and handling it is a challenging task in data stream mining. If an algorithm is not adapted to concept drift, then it directly affects its performance. A number of algorithms have been developed to handle concept drift, but they are not suited for both sudden concept drift and gradual concept drift. Thus, there is a demand for an algorithm that can react to both the types of concept drift as well as incur less computational cost. A new approach hybrid early drift detection method (HEDDM) has been proposed for drift detection, which works with an ensemble method to improve the performance.

Suicide is the fourth leading cause of death among people aged 15-29. More than 720, 000 people commit suicide every year. During the COVID-19 pandemic, we saw an increase in people seeking out mental health support on anonymous forums like Reddit. These anonymous forums allow people to express their suicidal ideation without judgment and give them a support structure that not everyone has. The aim of this project is to detect suicidal ideation using Reddit. In this work, we propose SIRGEL (Suicidal Ideation on Reddit using Graph Embeddings and LLMs), a dual-pipeline approach that combines large language model (LLM)- based annotation with both traditional text classification and graph neural network (GNN)-based modeling. The first step is to create a Reddit dataset for related subreddits. The second step is to annotate them as suicidal or not suicidal using LLMs. We also detail the creation of a new dataset made using Reddit posts and its benchmarking using baseline machine learning models. We model this problem using two approaches: traditional text classification and graph representation learning. Additionally, experiments with different types of Graph Neural Networks are performed with optimizations to improve and achieve efficient results

The application of deep learning to tabular data remains an ongoing challenge, with tree-based models such as XGBoost consistently outperforming neural network methods in most real-world scenarios. Recent advances in Large Language Models (LLMs) highlight their ability to generalize to new tasks via few-shot or one-shot prompting, yet questions remain about their utility in structured, tabular domains—particularly for resource-efficient Small Language Models (SLMs). In this work, we systematically evaluate the effectiveness of both large and small transformer-based language models for tabular classification tasks using few-shot strategies. Our approach investigates input serialization schemes and prompt engineering to maximize performance in low-data regimes. Experiments on benchmark datasets—including Diabetes, Heart Failure, and German Credit Risk—demonstrate that well-designed SLMs can approach, and occasionally match, the performance of much larger models, substantially reducing the need for extensive labeled data and retraining. However, further advances are required for language models to consistently rival the strongest tree-ensemble baselines. Our findings support the idea that SLMs, when properly prompted, offer a promising, flexible, and label-efficient alternative for automating and dynamizing machine learning pipelines on tabular data.

Large Language Models (LLMs) have tremendous potential to play a key role in supporting mathematical reasoning, with growing use in education and AI research.However, most existing benchmarks are limited to English, creating a significant gap for lowresource languages.For example, Bangla is spoken by nearly 250 million people who would collectively benefit from LLMs capable of native fluency.To address this, we present BanglaMATH, a dataset of 1.7k Bangla math word problems across topics such as Arithmetic, Algebra, Geometry, and Logical Reasoning, sourced from Bangla elementary school workbooks and annotated with details like grade level and number of reasoning steps.We have designed BanglaMATH to evaluate the mathematical capabilities of both commercial and open-source LLMs in Bangla, and we find that Gemini 2.5 Flash and DeepSeek V3 are the only models to achieve strong performance, with 80% accuracy across three elementary school grades.Furthermore, we assess the robustness and language bias of these top-performing LLMs by augmenting the original problems with distracting information, and translating the problems into English.We show that both LLMs fail to maintain robustness and exhibit significant performance bias in Bangla.Our study underlines current limitations of LLMs in handling arithmetic and mathematical reasoning in lowresource languages, and highlights the need for further research on multilingual and equitable mathematical understanding.

Medical benchmark datasets significantly contribute to developing Large Language Models (LLMs) for medical knowledge extraction, diagnosis, summarization, and other uses. Yet, current benchmarks are mainly derived from exam questions given to medical students or cases described in the medical literature, lacking the complexity of real-world patient cases that deviate from classic textbook abstractions. These include rare diseases, uncommon presentations of common diseases, and unexpected treatment responses. Here, we construct Clinically Uncommon Patient Cases and Diagnosis Dataset (CUPCase) based on 3,563 real-world case reports from BMC, which we formulate into diagnoses in open-ended textual format and as multiple-choice options with distractors. Using this dataset, we evaluate the ability of state-of-the-art LLMs, including both general-purpose and Clinical LLMs, to identify and correctly diagnose a patient case, and test models' performance when only partial information about cases is available. Our findings show that general-purpose GPT-4o attains the best performance in both the multiple-choice task (average accuracy of 87.9%) and the open-ended task (BERTScore F1 of 0.764), outperforming several LLMs with a focus on the medical domain such as Meditron-70B and MedLM-Large. Moreover, GPT-4o was able to maintain 87% and 88% of its performance with only the first 20% of tokens of the case presentation in multiple-choice and free text, respectively, highlighting the potential of LLMs to aid in early diagnosis in real-world cases. An error analysis demonstrates the complexity of the task, and attempts to hypothesise about the models' reasoning. CUPCase expands our ability to evaluate LLMs for clinical decision support in an open and reproducible manner.