POEM 2024: abstracts of contributed presentations

Kateřina Haniková: Explaining misleading claims using graphs of entities

Many researchers from various fields are focused on countering the spread of misinformation. In the literature, we can find two main strategies for reducing the spread of misinformation – debunking and prebunking (or inoculation). While the former is focused more on explaining misinformation after it occurs in public space, the latter is about teaching people to think critically and prevent them from sharing harmful content. The activity called fact-checking is focused on debunking. It is a complex process with several steps, starting from identifying a claim that should be verified and collecting evidence that supports or refutes the claim. After that, a verdict is stated, and a fact-checker writes a report that consists of argumentation supported by the evidence. Different fact-checkers use different scales of verdicts or typologies of misinformation, which can lead to confusion because there is no harmonization or unified methodology. An approach considered in this PhD research is focused on analyzing and classifying fact-checked misleading claims with the help of graph structures, which provides a new perspective on how to look at those claims. The aim is to develop a formal ontological model that would allow the capture of fact-checkers’ argumentation and support explaining misleading claims. There are two possible applications of this research. The first one, explaining misleading claims, is about explaining claims to users who would like to learn about misinformation or are interested in what is wrong with the claim and did not want to read a full fact-checked report or did not understand the report. The second application is about supporting fact-checking activity by training model that would suggest questions that should be checked.

Patrik Kompuš: Adopting ontology as a backbone of future-proof software development

The development of a software, in relation to business demands, is often very slow. Possible solution for this issue might be assigning ownership of the data model to data experts instead of development teams. The proposition is that ontologies are suitable means of supporting this separation. In this PhD project, I am attempting to partially verify whether a new methodology and approach based on ontologies will allow us to lower the number of resources spent on developing new products and future product features. These resources include time, personnel and costs. First use-case, where this ownership switch is being implemented, is in the processing of semantically enriched data. To improve the developers’ experience of using the Semantic Web serialization formats, e.g., RDF or Turtle, an intermediate solution might be beneficial. An artifact supporting dynamic serialization of ontology metadata graphs, the “possible” structure, with ontology mapping paths, the “required” structure was  implemented as a tool, RDF2JSON-OM, and is demonstrated as a web service endpoint.

Veronika Kostrouchová: Information models in healthcare: CAR T-cell therapy

My current work examines the integration of conceptual and process models to enhance the understanding and management of advanced cancer treatment, namely CAR T-cell therapy – a breakthrough in precision medicine that uses genetically engineered T-cell to target cancer. In the first year of my studies, I focused on the Ontological model, PURO model and MMABP model to present a multi-layered approach to encompass the complexities of CAR T-cell therapy. The Ontological Model, that was based on ontological principles from the Unified Foundational Ontology and builds upon established ontological modeling techniques, specifically using OntoUML, was designed to provide a high-level, structured representation of the domain, encompassing key concepts and their relationships. The PURO model, made using a web-based tool called PURO Modeler, facilitates detailed graphical ontology sketching, further refining conceptual insights. The next goal in my work is to expand the PURO model to achieve a more detailed representation of the domain to incorporate additional ontological distinctions and more complex relationships among entities. Based on insights gained from this planned expanded PURO model, the Ontological model will also be updated to reflect new insights uncovered through the detailed analysis in the PURO model, ensuring the Ontological model remains accurate and comprehensive. The final model, the Methodology for Modeling and Analysis of Business Processes model, provides a process-centric perspective, mapping the workflow stages and the potential outcomes in the CAR T-cell therapy treatment process. Together, these models can offer a comprehensive framework for analyzing treatment efficacy, optimizing workflow, even supporting clinical decision-making, thus paving the way for potential applications in automating and improving the process.

Dana Malcová: Ontology-Driven Agility: A New Framework for Sustainable Organizational Evolution

This presentation explores possibilities for ontology-based design of business capabilities for sustainable, context-aware organizational agility. The aim of the underlying research is to enable strategic design, development, and seamless evolution of adaptive capabilities enabling organisations to thrive in fast-paced, dynamic environments where disruptive changes are increasingly common. The focus is on a conscious, intentional approach to organizational evolution, particularly with regard to economic sustainability and responsiveness to complex challenges.
The research investigates an ontology-based method allowing organizations to meet the ever-changing demands of primarily investors, customers, and regulatory bodies more effectively and innovatively. Specifically, the study examines the intersection of organizational agility and sustainability, with applications in PhD-level research funding and in context of the circular economy.
Applying an ontological approach, this method aims to establish a structured framework that enables organizations not only to adapt to changes but also to strategically anticipate them. The framework seeks to foster a shared language and understanding within organizations, enhancing communication, decision-making, and agility in response to funding dynamics and sustainability challenges.
The talk will introduce foundational concepts of the ontology-driven approach, presenting initial findings and discussing its potential to enable organizations to navigate and succeed in an evolving landscape.

Vojtěch Svátek, Ondřej Zamazal: Overcoming the structural heterogeneity in OWL ontologies

When studying OWL ontologies arising in different domains, it was repeatedly reported that the (nearly) same content is, or can be, expressed using somewhat different constellations of constructs (aka logical-structural patterns). The specific contexts in which such heterogeneity is observed may correspond to ontology alignment, ontology transformation, or ontology design in multiple structural style variants. The talk will briefly survey the associated, past and ongoing, research activities at VSE.

Peter Vajdečka: Enhancing Ontology Property and Subclass Generation with Few-Shot Learning Using Large Language Models

Ontology engineering is a foundational aspect of knowledge representation in information systems, crucial for structuring knowledge and enabling
interoperability among diverse systems. Traditional ontology development is labor-intensive and requires significant expertise. This work explores the application of Large Language Models (LLMs), specifically gpt-4o, o1-mini, and o1-preview, in automating the generation of ontology properties and subclasses using two ontology design patterns: Object Property Chain Shortcutting and Subclass Enrichment.

We investigate the effectiveness of zero-shot versus few-shot learning approaches. Zero-shot learning allows models to generate outputs without prior examples, while few-shot learning involves providing a small number of high-quality examples to guide the model’s output. Our study leverages datasets from previous research, selecting examples where LLM outputs were better than human suggestions as few-shot prompts, and using challenging cases where initial outputs were worse or equal to human suggestions as evaluation inputs.
Our methodology includes crafting detailed prompts for both patterns and conducting experiments with the three models in both zero-shot and
few-shot settings. We evaluate the models based on their ability to produce outputs that match or surpass human-generated suggestions.

The results demonstrate that few-shot learning significantly enhances model performance. gpt-4o, in particular, consistently produced outputs
that not only matched but often exceeded human suggestions in clarity and adherence to ontology naming conventions. Few-shot learning enabled
the models to generate more precise, semantically appropriate ontology properties and subclasses, effectively reducing the reliance on human expertise. This study underscores the potential of LLMs in ontology development, showing that with minimal high-quality examples, they can achieve
or surpass human-level performance in generating ontology components. The findings have profound implications for the field of knowledge engineering, suggesting that integrating few-shot learning with LLMs could streamline ontology development, making it more efficient and scalable.