TRACKING HIVEMIND’S JOURNEYfrom inception to breakthroughs_

This paper explores the underexamined area of fine-grained emotion classification in app reviews, extending beyond the traditional focus on sentiment polarity (positive, negative, neutral). To capture the complexity of users’ affective responses, the study adapts Plutchik’s emotion taxonomy and introduces a structured annotation framework and dataset tailored to app reviews. Through an iterative human annotation process, the authors establish clear guidelines, highlight challenges in interpreting emotions, and assess the feasibility of automation with large language models (LLMs). The results show that LLMs substantially reduce manual annotation effort and achieve notable agreement with human annotators, though full automation remains difficult due to the nuanced nature of emotions. This work provides structured guidelines, an annotated dataset, and insights for building semi-automated pipelines, offering valuable contributions to opinion mining, requirements engineering, and user feedback analysis.

This paper investigates the potential of Multi-Agent Debate (MAD) strategies to enhance the performance of Large Language Model (LLM) agents in Requirements Engineering (RE) tasks. While prior research has focused on prompt engineering, fine-tuning, and retrieval-augmented generation, these methods often treat LLMs as isolated black boxes, relying on single-pass outputs with limited robustness and adaptability. Inspired by the way human debates improve accuracy by incorporating diverse perspectives, this study explores whether collaborative interactions among multiple LLM agents can yield similar benefits. We systematically analyze existing MAD strategies across different domains, identifying their key characteristics and developing a taxonomy of core attributes. Building on this foundation, we implement and evaluate a preliminary MAD-based framework for RE classification. The results demonstrate both the feasibility and potential advantages of applying MAD to RE, paving the way for more robust, adaptive, and accurate use of LLMs in engineering contexts.

This paper examines the current landscape of AI-powered software testing tools by systematically reviewing and classifying 56 commercially available solutions as of 2024. It analyses how these tools support different stages of the software testing process, ranging from test planning and test-case design to execution and maintenance, and highlights their potential to improve efficiency and effectiveness for test engineers. At the same time, the paper identifies key limitations, including false positives and insufficient contextual or domain understanding, which underscores the continued need for human oversight. The study argues that AI-assisted testing tools should be seen as complementary to human testers rather than fully autonomous solutions, with close human–AI collaboration remaining essential in the foreseeable future.

SciLaD is a novel, large-scale dataset of scientific language constructed entirely using open-source frameworks and publicly available data sources. It comprises a curated English split containing over 10 million scientific publications and a multilingual, unfiltered TEI XML split including more than 35 million publications. We also publish the extensible pipeline for generating SciLaD. The dataset construction and processing workflow demonstrates how open-source tools can enable large-scale, scientific data curation while maintaining high data quality. Finally, we pre-train a RoBERTa model on our dataset and evaluate it across a comprehensive set of benchmarks, achieving performance comparable to other scientific language models of similar size, validating the quality and utility of SciLaD. We publish the dataset and evaluation pipeline to promote reproducibility, transparency, and further research in natural scientific language processing and understanding, including scholarly document processing.

We present PolyglotQL, an open-source ETL (Extract, Transform, Load) pipeline for systematically creating multilingual text-to-SPARQL datasets, along with an accompanying framework for evaluating text-to-SPARQL generation models. PolyglotQL provides an extensible and modular architecture that aggregates, normalizes, and augments heterogeneous question–SPARQL pairs from established text-to-SPARQL datasets. With this pipeline, we automatically construct a bilingual English–German dataset featuring contextualized entity and relationship mappings as well as automatically translated and aligned question pairs. We also conduct an empirical evaluation using two multilingual open large language models under two distinct contextualization settings. The results show consistent performance improvements when explicit grounding information is provided, highlighting the benefits of structured context in multilingual semantic parsing.

This experience report reflects on the process of conducting a Multivocal Literature Review on the use of generative AI agents and multi-agent systems in project management within Agile software development. It examines the challenge of balancing timely research with the maturity of available evidence in a rapidly evolving field, where preprints and grey literature play a significant role. The report presents the research protocol, key methodological choices, and challenges encountered during the initial stages of the review, including the search process, study selection, and quality assessment. It also outlines practical lessons for maintaining rigour while working iteratively, with particular attention to construct stability, the treatment of preprints, dual-track quality assurance, and adaptations to the review process. These lessons may support researchers planning secondary studies in similarly fast-moving research areas.

Large Language Models (LLMs) depend on high-quality, domain-specific natural language datasets. This dependency is particularly pronounced in Requirements Engineering (RE), where core activities rely on textual artifacts such as requirements, specifications, and stakeholder feedback. Despite the increasing use of LLMs in RE, data scarcity remains a widely reported limitation. While several datasets support LLM-based RE research, they are scattered across studies and lack systematic characterization, hindering reuse, comparability and assessment. This paper addresses this gap by examining which public datasets are used in LLM-based RE, how they can be consistently characterized, and which RE tasks and dataset properties remain under-represented. We report on a systematic mapping study of 45 primary studies referencing 62 publicly available datasets. Each dataset is characterized using a structured scheme covering multiple dimensions, including relevant descriptors such as artifact type, granularity, RE activity, supported task, application domain, and language, among others. The results reveal notable imbalances, including an incomplete adoption of open-science practices, limited dataset support for elicitation activities, and a lack of language and socio-technical diversity. The resulting catalogue and characterisation scheme support informed dataset selection, comparison, and reuse, contributing to stronger empirical foundations for LLM-based RE research and evaluation.

Autonomous Driving Systems (ADS) can fail because of faults within individual modules as well as from interactions across perception, planning, and control. Yet existing ADS testing research often treats key testing functions, such as perturbation generation, behavioural assessment, and test case selection and exploration, as loosely coupled steps rather than coordinated roles for discovering such failures. We present CREAD, a collaborative multi-agent testing framework for testing ADS that organises perturbation generation, behavioural validation, and search coordination through a shared blackboard and an orchestrator. In the current work-in-progress instantiation, the framework focuses on perception-oriented perturbation generation, while remaining extensible to other ADS modules, including planning and control. It currently comprises a Perception Fuzzer Agent, a Metamorphic Validator Agent, and an Orchestrator Agent. Respectively, they generate perturbations, assess behavioural consistency across related scenario pairs, and coordinate further exploration. Experiments in HighwayEnv simulator show that the collaborative configuration improves failure discovery in the highway environment and remains competitive in the roundabout setting. Across the two environments, it yields about 2.1x as many failures per 100 scenarios as the single-agent baseline on average, while gains over a non-collaborative two-agent baseline vary across environments. These results suggest that collaborative multi-agent testing is a promising research direction for emergent ADS behaviour discovery.

Reinforcement learning (RL) agents often operate as black boxes, making it difficult to understand their decision-making in dynamic environments. This study proposes a novel framework for explainable RL based on structural causal models (SCMs). Here, the approach learns an SCM of the environment dynamics and reward process in a mobile network simulator (mobile-env), and uses this causal model to generate counterfactual explanations and perform interventions to understand agent behavior. The approach demonstrates that the learned SCM can closely approximate the environment’s transition dynamics while remaining interpretable. By leveraging do-calculus and counterfactual reasoning, our framework explains the long-term effects of actions through causal chains and highlights key influential factors. Experiments on a wireless network control task show that our method provides meaningful explanations for agent decisions (e.g., why a given action yields a higher reward), with minimal loss in policy performance. The study also presents comparative evaluations against baseline explanation approaches and discusses how our SCM-based explanations improve transparency and trust in RL policies.

Extrusion-based 3D printing processes of industrial plastic granulate expand the application possibilities of fused layer modeling (FLM) through higher deposition rates at lower material costs. This enables the economical production of large component structures, such as large tool molds or hull segments for ship constructions, which, however, places higher demands on process reliability and dimensional stability. Many parameters, such as the extruder speed, material quality or temperature, influence the shape of the extruded rope and therefore also the dimensional accuracy of the overall printed geometry. In addition, elementary influencing parameters are subject to process related fluctuations, which leads to imperfections and local deviations from the originally planned geometry. This results in the requirement to calculate and represent an exact knowledge of the actual geometry during the printing process in the one hand to localize defects and deviations and in the other hand to improve the actual printed geometry for the planning of finishing machining process. This article presents a tri-dexel-based simulation model that uses material data, extruder data and machine control data, which are acquired in a high-frequency interpolation cycle during the process, to calculate the material deposition and the resulting actual geometry and make it available as a CAD model. With this new method, complex geometric three-dimensional component measurements, e.g. with a laser line scanner, can be reduced and even substituted. The use of this actual geometry enables an adapted parameter and design optimization in the planning and design of extrusion-based 3D printing processes. In addition, this calculation method enables the generation of a complete digital twin in which the geometry with all its relevant properties is mapped in parallel to the physical product.