October 25 ~ 26, 2025, Vienna, Austria
Davide AVESANI, Ammar KHEIRBEK, Isep - Institut Sup´erieur d’Electronique de Paris ´10 rue de Vanves, 92130 Issy-les-Moulineaux, France
Social media is now deeply integrated into people’s daily life, enabling rapid information exchange and global connectivity. Unfortunately, harmful content can be easily disseminated among all communities, including hate speech and biases against vulnerable groups such as people with disabilities. While social media platforms employ a mix of automated systems and skillful experts for content moderation, significant challenges remain in detecting nuanced hate speech, particularly when expressed through indirect or coded language. This paper proposes a novel approach to address these challenges through HEROL (Hate-speech Evaluation via RAG and Optimized LLM), a unified model that combines RAG-Enhanced Large Language Models with Prompt Engineering and Fine-Tuning. Experimental results, obtained through a structured evaluation methodology using annotated social media datasets, demonstrated that HEROL achieved an accuracy improvement by up to 10% compared to baseline models. This highlights its effectiveness in identifying subtle and indirect forms of hate speech and its potential to contribute to safer, more inclusive online environments.
Social Media – Hate Speech Detection – Disability – Natural Language Processing – Large Language Models – Prompt Engineering – Fine-Tuning – Retrieval-Augmented Generation – Knowledge Graph
Liya Wang, David Yi,Damien Jose,John Passarelli, James Gao, Jordan Leventis, and Kang Li, Atlassian, USA
Large Language Models (LLMs) enhance productivity through AI tools, yet existing benchmarks like Multitask Language Understanding (MMLU) inadequately assess enterprise-specific task complexities. We propose a 14-task framework grounded in Bloom’s Taxonomy to holistically evaluate LLM capabilities in enterprise contexts. To address challenges of noisy data and costly annotation, we develop a scalable pipeline combining LLM-as-a-Labeler, LLM-as-a-Judge, and corrective retrieval-augmented generation (CRAG), curating a robust 9,700-sample benchmark. Evaluation of six leading models shows open-source contenders like DeepSeek R1 rival proprietary models in reasoning tasks but lag in judgment-based scenarios, likely due to overthinking. Our benchmark reveals critical enterprise performance gaps and offers actionable insights for model optimization. This work provides enterprises a blueprint for tailored evaluations and advances practical LLM deployment.
Large Language Models (LLMs), Evaluation Benchmark, Bloom’s Taxonomy, LLM-as-a-Labeler, LLM-as-a-Judge, corrective retrieval-augmented generation (CRAG).
Philipp Seitz, Jan Schmitt, and Andreas Schiffler, Institute of Digital Engineering, Technical University of Applied Sciences W¨urzburg- Schweinfurt, Germany
For a larger set of predictions of several differently trained machine learning models, known as bagging predictors, the mean of all predictions is taken by default. Nevertheless, this proceeding can deviate from the actual ground truth in certain parameter regions. A method is presented to determine a representative value ˜yBS from such a set of predictions and to evaluate it by an associated quality criterion βBS, called Bagging Score (BS), using nonlinear regression with Neural Networks (NN). The BS reflects the confidence of the obtained ensemble prediction and also allows the construction of a prediction estimation function δ(β) for specifying deviations that are more precise than using the variance of the bagged predictors themselves.
Machine Learning, Neural Network, Bagging Predictors, Bagging Score, Nonlinear Regression, Deviation Estimation.
Stepan Plotytsia, Delivery Manager, Grid Dynamics Holdings, Inc., Schaumburg, Illinois, USA
This research presents a comprehensive proof of concept (PoC) study comparing migration strategies from monolithic SAP Commerce platforms to cloud-native microservices architectures. I propose an innovative six-phase transformation methodology that incorporates composable commerce patterns, AI-driven personalization, and event-first integration. Through simulation modelling, theoretical analysis, and a comparative evaluation against alternative approaches (modular monolith, lift-and-shift, and phased SOA), I project potential improvements of 60% in latency reduction, 4× deployment frequency, and 282% ROI over five years. The study employs mixed-methods research, combining quantitative modelling with qualitative analysis of organizational readiness factors. This framework includes detailed architectural blueprints, comprehensive risk mitigation strategies, implementation roadmaps, and decision matrices validated through industry benchmarks and theoretical modelling. This research aims to provide organizations with data-driven insights and actionable guidance for evaluating modernization strategies before committing to full-scale transformation, addressing the critical gap in empirical comparison of migration approaches for enterprise e-commerce platforms.
Proof of concept, Migration strategy comparison, SAP Commerce modernization, Microservices architecture, ROI modelling, Risk assessment, Cloud transformation, Composable commerce, Event-driven architecture, Digital transformation
Nasreddine Aoumeur1, Kamel Barkaoui2, Gunter Saake3, 1University of Science and Technology (USTO), Algeria, 2Laboratoire CEDRIC, CNAM, 292 Saint Martin, 75003 Paris - FRANCE, 3ITI, FIN, Otto-von-Guericke-Universit¨at Magdeburg, Germany
Whereas Artificial Intelligence (AI) with its Machine-Learning (ML) vertiginous advances are significantly reshaping our way of developing software either as (prompting) GenerativeAI- or purely ML-based ones, any significant influence of decades of investigations and findings around software-engineering (SE) concepts, principles and methods on such new AI-Era software is unfortunately almost desperately missing. The resulting is plethora of GenerativeAI- and MLbased software: Black-boxed rigid ill-conceptually and completely isolated from our ”ordinary” yet mostly disciplined software landscape. The aim of this paper is to contribute in leveraging such unsatisfactory Promptingand ML-based software form to be well-conceptually, dynamically adaptable by intrinsically fitting it within our ”ordinary” domain-oriented software landscape: We refer generically to as AI-Powered (knowledge-intensive) applications software; thereby reconciliating Domain- and AI-Experts instead of contemporarily miserable ’confrontation’. We achieve such promising endeavour by exactly capitalizing on best advanced SE concepts and principles More precisely, we are putting forward an innovative stepwise integrated modeldriven approach that smoothly exhibits the following conceptual, founded and technological milestones. Firstly, any structural features are semi-formally modelled as UML components intrinsically thereafter mapped into (ordinary and MLbased) Web-Services. Behavioural crucial features are then captured as intuitive business rules mostly at the inter-service interactions. Secondly, for the precise conceptualization of such inter-service behavioural rules, we are proposing tailored graphically appealing stereotyped primitives as ECA-driven architectural (interservice) connector glues, we refer to as ECA-driven interaction laws. Thirdly, for the rigorous certification, while staying ECA-Compliant we are tailoring Meseguer’s true-concurrent rewriting logic and its strategies-enabled Maude language for that purpose. Last but not least, for the efficient implementation we are proposing a four-level implementation still ECA-Compliant architecture, by relying on modern software technologies including python-empowered API with Django and its REST framework and Visual-Studio enterprise as advanced IDE. All approach milestones and steps are extensively illustrated using a quite realistic AI-powered software application dealing with Brain Tumor diagnostics while stressing all its benefits, with at-top reliability, dynamic-adaptability, self-learning and understandability.
ECA-driven architectural interaction laws, UML and Service-orientation, Machine-Learning (ML), KNN, Brain-Tumor, Reliability and Adaptability, RewritingLogic, Domain- and AI-Experts, Django REST and Python API.
Luz-Viviana Cobaleda-Estepa1, Julián Carvajal2, Paola Vallejo3, Andrés López4, Raúl Mazo5, 1Facultad de Ingeniería, Universidad de Antioquia, Medellín, Colombia, 2Facultad de Ingeniería, Universidad de Antioquia, Medellín, Colombia, 3Escuela de Ciencias Aplicadas e Ingeniería, Universidad EAFIT, Medellín, Colombia, 4Facultad de Ingeniería - Universidad de Antioquia, Medellín, Colombia, 5Lab-STICC, ENSTA, Brest, Francia.
Modern software systems increasingly integrate machine learning (ML) due to its advancements and ability to enhance data-driven decision-making. However, this integration introduces significant challenges for software engineering, especially in software product lines (SPLs), where managing variability and reuse becomes more complex with the inclusion of ML components. Although existing approaches have addressed variability management in SPLs and the integration of ML components in isolated systems, few have explored the intersection of both domains. Specifically, there is limited support for modeling and managing variability in SPLs that incorporate ML components. This article addresses this gap by proposing a structured framework that enhances SPL to support the inclusion of ML components. It facilitates the design of SPLs with ML capabilities by enabling systematic modeling of variability and reuse. The proposal has been partially implemented with the VariaMos tool.
Machine Learning (ML), Software Product Lines (SPL), ML-based systems, variability modeling.
Tanay Kamlesh Patel, Niraj Anil Babarl, Deep Pujaral, Glen Ueharal, JeanLarson2, Andreas Spanias2, lSenSIP Center, Schoo of ECEE, Arizona State University, Tempe, USA, 2 USA
This study describes the development of the Research Experience for Undergraduate students (REU) training program on Quantum Machine Learning (QML), hosted by the SenSIP Center at Arizona State University. In 2025, the REU hosted several projects that engaged quantum computing in signal processing, audio analysis, computer vision, medical diagnostics, anomaly detection, and generative machine learning. The objectives of the training program are to a) engage students in QML research by immersing them in government and industry projects, b) train students in quantum information processing and machine learning simulations, c) encourage students to pursue graduate research, d) increase awareness of career opportunities in QML, and e) provide professional development training. As part of professional development, students presented to stakeholders and received training in preparing publications, building awareness on social implications, ethics, and privacy. The program is evaluated by the Center for Evaluating the Research Pipeline (CERP) and an independent evaluator. This paper describes the importance of introducing QML research at the undergraduate level, recruitment, program structure, summaries of REU projects, and preliminary evaluations.
REU, Quantum Computing, Quantum Machine Learning, Qubits, Workforce Development
Osama S. Qatrani, Independent Researcher, UK
This paper introduces a symbolic governance model that maps how political authorities (P1–P2, L, I), capital (F, C), and layered technical infrastructures (T1–T4) translate directives into recursive algorithmic control. Rather than treating AI as an autonomous or neutral technology, the model reframes it as an encoded structure of power operating across data (D), social interfaces (S1–S2), and global arenas (G1–G3). By compressing complex system interactions into an accessible symbolic language, the framework helps non-technical stakeholders trace influence from policy and finance to code, platforms, and behavioral feedback (R). Brief case snapshots illustrate how algorithmic logics shape information, decisions, and public life. The contribution is a practical lens for diagnosing power in AI ecosystems and a policy-oriented roadmap for oversight, transparency in optimization targets, and public-interest safeguards against algorithmic domination.
Artificial Intelligence; Algorithmic Governance; Political Power; Digital Politics; Algorithmic Control.