Anomaly-based ML-NIDS (A-NIDS) model normal network behavior from benign data and classify deviations from this baseline as anomalies, theoretically enabling the detection of evolving attack variants without labeled attack data. The ability of A-NIDS to generalize critically depends on the quality of the feature space representing network behavior. However, the requirement for feature spaces that encode attack-relevant semantics has received little attention and remains poorly understood. As a consequence, these systems still struggle to meet practical operational constraints (low false positive rates without compromising detection performance and generalization to attack variants). We identify two limitations in the current feature spaces. First, Out-of-Dimension Blindness, where features do not capture essential attack mechanism properties. Second, Attack Strategy Aggregation Failure, where features cannot encode composite attack behaviors. Moreover, we demonstrate that two SotA data-driven generalization frameworks (based on incremental and contrastive learning) cannot compensate for these feature-level shortcomings. To bridge this gap, we present KnowML, a framework that encodes attack domain knowledge directly into the feature space. For each attack family, our method employs LLMs to construct a corresponding Knowledge Graph (KG) from attack implementations. Symbolic reasoning is then applied over the KG to enumerate potential attack strategies and their compositions. The resulting Knowledge-Augmented Feature Space enables effective generalization even when trained exclusively on benign traffic, a capability beyond current approaches. Systematic empirical evaluations show that KnowML achieves up to 99% detection rates while maintaining false positive rates at or below 0.0137%, substantially outperforming contemporary feature-based baselines across diverse attack variants.

An ontology is a formal, explicit specification of a shared conceptualization that, with problem‑solving and reasoning methods, supports efficient semantic technology development. In ontology engineering, Competency Questions (CQs) capture functional requirements that define an ontology’s application domain. Auditing this domain scope with CQs is challenging because in nature, there are no clear domain boundaries, and ontology engineers must then decide which subdomains to cover (horizontal coverage) and how much detail to model (vertical granularity) in an ontology. LLM‑based systems can generate many candidate CQs to guide these decisions, but current tools underuse this potential: they lack support for users’ divergent (lateral) and convergent (vertical) thinking in a visualized CQs space organized by coverage and granularity. As a result, users struggle to systematically decide which CQs to adopt, discard, or refine. We propose an interaction framework that fills this gap, demonstrated through OntoScope, an LLM‑based interactive system, and a user study with 15 ontology engineers. To our knowledge, this is the first validated interaction framework with an LLM‑based system that helps ontology engineers audit domain boundaries and unifies fragmented, expert‑driven ontology scoping practices into a coherent, accessible approach. More broadly, it shows how LLM‑based systems can transparently and accountably support a wider range of knowledge‑intensive tasks.

An ontology is a formal, explicit specification of a shared conceptualisation, which can be combined with problem-solving methods and reasoning functionality to develop high-quality technology and application systems efficiently. Ontology engineering typically involves extensive manual effort to elicit intended use cases (user stories) from users for the target ontology-based systems. Recent studies have demonstrated the positive potential of large language model-based conversational agents in supporting user story generation in OE. However, we argue that we are not leveraging LLM to its fullest potential by not supporting users in formulating effective prompts. To address this, we identify the prompt guidance users need during user story generation workflows by conducting a formative study (N = 10) using participatory prompting. We demonstrate its usefulness through the design and development of the OntoChat LLM-based system for OE, as well as a user evaluation with knowledge engineers (N = 24). To our knowledge, this is the first work to design and validate a prompt guidance framework that helps users leverage LLM to its fullest potential to generate effective requirements for ontology development. This advances how we interact with LLM for requirements elicitation.

Despite many advances in knowledge engineering (KE), challenges remain in areas such as engineering knowledge graphs (KGs) at scale, automating tasks, and keeping pace with evolving domain knowledge. KE has used NLP demonstrating notable advantages in knowledge-intensive tasks, but the most effective use of generative AI to support knowledge engineers across the KE activities is still in its infancy. To explore how generative AI may enhance KE and change existing KE practices, we conducted a multi-method study during a KE hackathon. We investigated participants’ views on the use of generative AI, the challenges they face, the skills they may need to integrate generative AI into their practices, and how they use generative AI responsibly. We found participants felt LLMs could indeed contribute to improving efficiency when engineering KGs, but presented increased challenges around the already complex issues of evaluating KE task success. We discovered prompting to be a useful but undervalued skill for knowledge engineers working with LLMs, and note that NLP skills may become more relevant across more roles in KE workflows. Integrating generative AI into KE tasks needs to be done with awareness of potential risks and harms. Given the limited ethical training most knowledge engineers receive, solutions such as our proposed ‘KG Cards’ based on Data Cards could be a useful guide for KG construction. Our findings can support designers of KE AI copilots, KE researchers, and practitioners using advanced AI to develop trustworthy applications, propose new methodologies for KE and operate new technologies responsibly.

Abusive language detection relies on understanding different levels of intensity, expressiveness and targeted groups, which requires commonsense reasoning, world knowledge and linguistic nuances that evolve over time. Here, we frame the problem as a knowledge-guided learning task, and demonstrate that LLMs’ implicit knowledge without an accurate strategy is not suitable for multi-class detection nor explanation generation. We publicly release GLlama Alarm, the knowledge-Guided version of Llama-2 instruction fine-tuned for multi-class abusive language detection and explanation generation. By being fine-tuned on structured explanations and external reliable knowledge sources, our model mitigates bias and generates explanations that are relevant to the text and coherent with human reasoning, with an average 48.76% better alignment with human judgment according to our expert survey.

This paper introduces DSrepair, a knowledge-enhanced program repair approach designed to repair the buggy code generated by LLMs in the data science domain. DSrepair uses knowledge graph based RAG for API knowledge retrieval and bug knowledge enrichment to construct repair prompts for LLMs. Specifically, to enable knowledge graph-based API retrieval, we construct DS-KG (Data Science Knowledge Graph) for widely used data science libraries. For bug knowledge enrichment, we employ an abstract syntax tree (AST) to localize errors at the AST node level. We evaluate DSrepair's effectiveness against five state-of-the-art LLM-based repair baselines using four advanced LLMs on the DS-1000 dataset. The results show that DSrepair outperforms all five baselines. Specifically, when compared to the second-best baseline, DSrepair achieves substantial improvements, fixing 44.4%, 14.2%, 20.6%, and 32.1% more buggy code snippets for each of the four evaluated LLMs, respectively. Additionally, it achieves greater efficiency, reducing the number of tokens required per code task by 17.49%, 34.24%, 24.71%, and 17.59%, respectively.

In this work, we study disagreements in discussions around Wikidata, an online knowledge community that builds the data backend of Wikipedia. Discussions are essential in collaborative work as they can increase contributor performance and encourage the emergence of shared norms and practices. While disagreements can play a productive role in discussions, they can also lead to conflicts and controversies, which impact contributor’ well-being and their motivation to engage. We want to understand if and when such phenomena arise in Wikidata, using a mix of quantitative and qualitative analyses to identify the types of topics people disagree about, the most common patterns of interaction, and roles people play when arguing for or against an issue. We find that decisions to create Wikidata properties are much faster than those to delete properties and that more than half of controversial discussions do not lead to consensus. Our analysis suggests that Wikidata is an inclusive community, considering different opinions when making decisions, and that conflict and vandalism are rare in discussions. At the same time, while one-fourth of the editors participating in controversial discussions contribute legitimate and insightful opinions about Wikidata’s emerging issues, they respond with one or two posts and do not remain engaged in the discussions to reach consensus. Our work contributes to the analysis of collaborative KG construction with insights about communication and decision-making in projects, as well as with methodological directions and open datasets. We hope our findings will help managers and designers support community decision-making and improve discussion tools and practices.

As (generative) Artificial Intelligence continues to evolve, so do the challenges associated with governing the data that powers it. Ensuring data quality, privacy, security, and ethical use become more and more challenging due to the increasing volume and variety of the data, the complexity of AI models, and the rapid pace of technological advancement. Knowledge graphs have the potential to play a significant role in enabling data governance in AI, as we move beyond their traditional use as data organisational systems. To address this, we present KG.GOV, a framework that positions KGs at a higher abstraction level within AI workflows, and enables them as a backbone of AI data governance. We illustrate the three dimensions of KG.GOV: modelling data, alternative representations, and describing behaviour; and describe the insights and challenges of three use cases implementing them: Croissant, a vocabulary to model and document ML datasets; WikiPrompts, a collaborative KG of prompts and prompt workflows to study their behaviour at scale; and Multimodal transformations, an approach for multimodal KGs harmonisation and completion aiming at broadening access to knowledge.

The process of developing ontologies – a formal, explicit specification of a shared conceptualisation – is addressed by well-known methodologies. As for any engineering development, its fundamental basis is the collection of requirements, which includes the elicitation of competency questions. Competency questions are defined through interacting with domain and application experts or by investigating existing datasets that may be used to populate the ontology i.e. its knowledge graph. The rise in popularity and accessibility of knowledge graphs provides an opportunity to support this phase with automatic tools. In this work, we explore the possibility of extracting competency questions from a knowledge graph. This reverses the traditional workflow in which knowledge graphs are built from ontologies, which in turn are engineered from competency questions. We describe in detail RevOnt, an approach that extracts and abstracts triples from a knowledge graph, generates questions based on triple verbalisations, and filters the resulting questions to yield a meaningful set of competency questions; the WDV dataset. This approach is implemented utilising the Wikidata knowledge graph as a use case, and contributes a set of core competency questions from 20 domains present in the WDV dataset. To evaluate RevOnt, we contribute a new dataset of manually-annotated high-quality competency questions, and compare the extracted competency questions by calculating their BLEU score against the human references. The results for the abstraction and question generation components of the approach show good to high quality. Meanwhile, the accuracy of the filtering component is above 86%, which is comparable to the state-of-the-art classifications.

Abstract In the music domain, several ontologies have been proposed to annotate musical data, in both symbolic and audio form, and generate semantically rich Music Knowledge Graphs. However, current models lack interoperability and are insufficient for representing music history and the cultural heritage context in which it was generated; risking the propagation of recency and cultural biases to downstream applications. In this article, we propose the Polifonia Ontology Network (PON) for music cultural heritage, centred around four modules: Music Meta (metadata), Representation (content), Source (provenance) and Instrument (cultural objects). We design PON with a strong accent on cultural stakeholder requirements and competency questions (CQs), contributing an NLP-based toolkit to support knowledge engineers in generating, validating, and analysing them; and a novel, high-quality CQ dataset produced as a result. We show current and future use of these resources by internal project pilots, early adopters in the music industry, and opportunities for the Semantic Web and Music Information Retrieval communities.

Computationally creative systems for music have recently achieved impressive results, fuelled by progress in generative machine learning. However, black-box approaches have raised fundamental concerns for ethics, accountability, explainability, and musical plausibility. To enable trustworthy machine creativity, we introduce the Harmonic Memory, a Knowledge Graph (KG) of harmonic patterns extracted from a large and heterogeneous musical corpus. By leveraging a cognitive model of tonal harmony, chord progressions are segmented into meaningful structures, and patterns emerge from their comparison via harmonic similarity. Akin to a music memory, the KG holds temporal connections between consecutive patterns, as well as salient similarity relationships. After demonstrating the validity of our choices, we provide examples of how this design enables novel pathways for combinational creativity. The memory provides a fully accountable and explainable framework to inspire and support creative professionals – allowing for the discovery of progressions consistent with given criteria, the recomposition of harmonic sections, but also the co-creation of new progressions.

Web vocabularies (WV) have become a fundamental tool for structuring Web data: over 10 million sites use structured data formats and ontologies to markup content. Maintaining these vocabularies and keeping up with their changes are manual tasks with very limited automated support, impacting both publishers and users. Existing work shows that machine learning can be used to reliably predict vocabulary changes, but on specific domains (e.g. biomedicine) and with limited explanations on the impact of changes (e.g. their type, frequency, etc.). In this paper, we describe a framework that uses various supervised learning models to learn and predict changes in versioned vocabularies, independent of their domain. Using well-established results in ontology evolution we extract domain-agnostic and human-interpretable features and explain their influence on change predictability. Applying our method on 139 WV from 9 different domains, we find that ontology structural and instance data, the number of versions, and the release frequency highly correlate with predictability of change. These results can pave the way towards integrating predictive models into knowledge engineering practices and methods.

Linked lists represent a countable number of ordered values , and are among the most important abstract data types in computer science. With the advent of RDF as a highly expressive knowledge representation language for the Web, various implementations for RDF lists have been proposed. Yet, there is no benchmark so far dedicated to evaluate the performance of triple stores and SPARQL query engines on dealing with ordered linked data. Moreover, essential tasks for evaluating RDF lists, like generating datasets containing RDF lists of various sizes, or generating the same RDF list using different modelling choices, are cumbersome and unprincipled. In this paper, we propose List.MID , a systematic benchmark for evaluating systems serving RDF lists. List.MID consists of a dataset generator, which creates RDF list data in various models and of different sizes; and a set of SPARQL queries. The RDF list data is coherently generated from a large, community-curated base collection of Web MIDI files, rich in lists of musical events of arbitrary length. We describe the List.MID benchmark, and discuss its impact and adoption, reusability, design, and availability.

Nanopublications are a Linked Data format for scholarly data publishing that has received considerable uptake in the last few years. In contrast to the common Linked Data publishing practice, nanopublications work at the granular level of atomic information snippets and provide a consistent container format to attach provenance and metadata at this atomic level. While the nanopublications format is domain-independent, the datasets that have become available in this format are mostly from Life Science domains, including data about diseases, genes, proteins, drugs, biological pathways, and biotic interactions. More than 10 million such nanopublications have been published, which now form a valuable resource for studies on the domain level of the given Life Science domains as well as on the more technical levels of provenance modeling and heterogeneous Linked Data. We provide here an overview of this combined nanopublication dataset, show the results of some overarching analyses, and describe how it can be accessed and queried.