Ontology Learning Research Articles

Dynamic Retrieval Augmented Generation of Ontologies using Artificial Intelligence (DRAGON-AI)

BackgroundOntologies are fundamental components of informatics infrastructure in domains such as biomedical, environmental, and food sciences, representing consensus knowledge in an accurate and computable form. However, their construction and maintenance demand substantial resources and necessitate substantial collaboration between domain experts, curators, and ontology experts.We present Dynamic Retrieval Augmented Generation of Ontologies using AI (DRAGON-AI), an ontology generation method employing Large Language Models (LLMs) and Retrieval Augmented Generation (RAG). DRAGON-AI can generate textual and logical ontology components, drawing from existing knowledge in multiple ontologies and unstructured text sources.ResultsWe assessed performance of DRAGON-AI on de novo term construction across ten diverse ontologies, making use of extensive manual evaluation of results. Our method has high precision for relationship generation, but has slightly lower precision than from logic-based reasoning. Our method is also able to generate definitions deemed acceptable by expert evaluators, but these scored worse than human-authored definitions. Notably, evaluators with the highest level of confidence in a domain were better able to discern flaws in AI-generated definitions. We also demonstrated the ability of DRAGON-AI to incorporate natural language instructions in the form of GitHub issues.ConclusionsThese findings suggest DRAGON-AI's potential to substantially aid the manual ontology construction process. However, our results also underscore the importance of having expert curators and ontology editors drive the ontology generation process.

LLMs4OL 2024 Datasets: Toward Ontology Learning with Large Language Models

Ontology learning (OL) from unstructured data has evolved significantly, with recent advancements integrating large language models (LLMs) to enhance various aspects of the process. The paper introduces the LLMs4OL 2024 datasets, developed to benchmark and advance research in OL using LLMs. The LLMs4OL 2024 dataset as a key component of the LLMs4OL Challenge, targets three primary OL tasks: Term Typing, Taxonomy Discovery, and Non-Taxonomic Relation Extraction. It encompasses seven domains, i.e. lexosemantics and biological functions, offering a comprehensive resource for evaluating LLM-based OL approaches Each task within the dataset is carefully crafted to facilitate both Few-Shot (FS) and Zero-Shot (ZS) evaluation scenarios, allowing for robust assessment of model performance across different knowledge domains to address a critical gap in the field by offering standardized benchmarks for fair comparison for evaluating LLM applications in OL.

Preface for LLMs4OL 2024: The 1st Large Language Models for Ontology Learning Challenge at the 23rd ISWC

We are pleased to present the proceedings of the “1st Large Language Models for Ontology Learning Challenge (LLMs4OL 2024)”, held at the 23rd International Semantic Web Conference (ISWC). This challenge marks a significant a dvancement in utilizing Large Language Models (LLMs) for Ontology Learning (OL)—a key Semantic Web component that facilitates the automatic extraction of structured knowledge from unstructured data. The challenge features three main tasks: Term Typing (identifying categories for terms), Taxonomy Discovery (uncovering hierarchical relationships), and Non-Taxonomic Relation Extraction (identifying other meaningful relationships between terms). Each task is designed to test different facets of ontology construction and to encourage the exploration of innovative techniques. This challenge seeks to foster collaboration, inspire innovation, and expand the capabilities of LLMs in OL. The proceedings include a collection of innovative solutions and insights from global participants, highlighting the crucial role of LLMs in enhancing the web with structured knowledge. We believe the outcomes of this challenge will propel further advancements in OL and its applications on the semantic web. We would like to extend our gratitude to all the participants for their invaluable contributions, which their solutions and dedication have greatly enriched this challenge. Our sincere thanks also go to the conference organizers and committee for their efforts in hosting this event. We are deeply appreciative of the reviewers for their evaluations and feedback. Their reviews have been instrumental in enhancing the quality of the submissions. We would like to specifically acknowledge: Dr. Amin Keshavarzi (Postdoctoral Researcher, L3S & TIB, Germany) Mostafa Rahgouy (Lead Project Coordinator, Auburn University, USA) Sahar Tahmasebi (Doctoral Researcher, TIB, Germany) Milad Molazadeh Oskuee (Lead NLP Researcher, Iran) Aida Usmanova (Doctoral Researcher, Leuphana Universit ¨ at L¨uneburg, Germany) Emetis Niazmand (Doctoral Researcher, TIB, Germany) Lastly, we would like to express our gratitude to ”Xenia Felice van Edig” and ”Karolina Linerová” from the TIB Open Publishing, for their support in bringing these proceedings online. We also would like to acknowledge that the 1st LLMs4OL Challenge @ ISWC 2024 jointly supported by the NFDI4DataScience initiative (DFG, German Research Foundation, Grant ID: 460234259) and the SCINEXT project (BMBF, German Federal Ministry of Education and Research, Grant ID: 01lS22070).

Phoenixes at LLMs4OL 2024 Tasks A, B, and C: Retrieval Augmented Generation for Ontology Learning

Large language models (LLMs) showed great capabilities in ontology learning (OL) where they automatically extract knowledge from text. In this paper, we proposed a Retrieval Augmented Generation (RAG) formulation for three different tasks of ontology learning defined in the LLMs4OL Challenge at ISWC 2024. For task A - term typing - we considered terms as a query and encoded the query through the Query Encoder model for searching through knowledge base embedding of types embeddings obtained through Context Encoder. Next, using Zero-Shot Prompt template we asked LLM to determine what types are appropriate for a given term within the term typing task. Similarly, for Task B, we calculated the similarity matrix using an encoder-based transformer model, and by applying the similarity threshold we considered only similar pairs to query LLM to identify whatever pairs have the "is-a" relation between a given type and in a case of having the relationships which one is "parent" and which one is "child". In final, for Task C -- non-taxonomic relationship extraction -- we combined both approaches for Task A and B, where first using Task B formulation, child-parents are identified then using Task A, we assigned them an appropriate relationship. For the LLMs4OL challenge, we experimented with the proposed framework over 5 subtasks of Task A, all subtasks of Task B, and one subtask of Task C using Mistral-7B LLM.

Silp_nlp at LLMs4OL 2024 Tasks A, B, and C: Ontology Learning through Prompts with LLMs

Our team, silp_nlp, participated in the LLMs4OL Challenge at ISWC 2024, engaging in all three tasks focused on ontology generation. The tasks include predicting the type of a given term, extracting a hierarchical taxonomy between two terms, and extracting non-taxonomy relations between two terms. To accomplish these tasks, we used machine learning models such as random forest, logistic regression and generative models for the first task and generative models such as llama-3-8b-instruct, mistral 8*7b and GPT-4o-mini for the second and third tasks. Our results showed that generative models performed better for certain domains, such as subtasks A6 and B2. However, for other domains, the prompt-based technique failed to generate promising results. Our team achieved first place in six subtasks and second place in five subtasks, demonstrating our expertise in ontology generation.

LLMs4OL 2024 Overview: The 1st Large Language Models for Ontology Learning Challenge

This paper outlines the LLMs4OL 2024, the first edition of the Large Language Models for Ontology Learning Challenge. LLMs4OL is a community development initiative collocated with the 23rd International Semantic Web Conference (ISWC) to explore the potential of Large Language Models (LLMs) in Ontology Learning (OL), a vital process for enhancing the web with structured knowledge to improve interoperability. By leveraging LLMs, the challenge aims to advance understanding and innovation in OL, aligning with the goals of the Semantic Web to create a more intelligent and user-friendly web. In this paper, we give an overview of the 2024 edition of the LLMs4OL challenge and summarize the contributions.

DaSeLab at LLMs4OL 2024 Task A: Towards Term Typing in Ontology Learning

The report presents the evaluation results of our approach in the LLM4OL Challenge, where we fine-tuned GPT-3.5 for Task A (Term Typing) across three different datasets. Our approach demonstrated consistent and robust performance during few- shot testing, achieving top rankings in several datasets and sub-datasets, proving the potential of fine-tuning LLMs for ontology creation tasks.

The Ghost at LLMs4OL 2024 Task A: Prompt-Tuning-Based Large Language Models for Term Typing

The LLMs4OL Challenge @ ISWC 2024 aims to explore the intersection of Large Language Models (LLMs) and Ontology Learning (OL) through three main tasks: 1) Term Typing, 2) Taxonomy Discovery and 3) Non-Taxonomic Relation Extraction. In this paper, we present our system's design for the term typing task. Our approach utilizes automatic prompt generation using soft prompts to enhance term typing accuracy and efficiency. We conducted experiments on several datasets, including WordNet, UMLS, GeoNames, NCI, MEDCIN, and SNOMEDCT_US. Our approach outperformed the baselines on most datasets, except for GeoNames, where it faced challenges due to the complexity and specificity of this domain, resulting in substantially lower scores. Additionally, we report the overall results of our approach in this challenge, which highlight its promise while also indicating areas for further improvement.

RWTH-DBIS at LLMs4OL 2024 Tasks A and B

The increasing capabilities of Large Language Models (LLMs) have opened new opportunities for enhancing Ontology Learning (OL), a process crucial for structuring domain knowledge in a machine-readable format. This paper reports on the participation of the RWTH-DBIS team in the LLMs4OL Challenge at ISWC 2024, addressing two primary tasks: term typing and taxonomy discovery. We used LLaMA-3-8B and GPT-3.5-Turbo models to find the performance gaps between open-source and commercial LLMs. For open-source LLMs, our methods included domain-specific continual training, fine-tuning, and knowledge-enhanced prompt-tuning. These approaches were evaluated on the benchmark datasets from the challenge, i.e., GeoNames, UMLS, Schema.org, and the Gene Ontology (GO), among others. The results indicate that domain-specific continual training followed by task-specific fine-tuning enhances the performance of open-source LLMs in these tasks. However, performance gaps remain when compared to commercial LLMs. Additionally, the developed prompting strategies demonstrate substantial utility. This research highlights the potential of LLMs to automate and improve the OL process, offering insights into effective methodologies for future developments in this field.

DSTI at LLMs4OL 2024 Task A: Intrinsic Versus Extrinsic Knowledge for Type Classification

We introduce semantic towers, an extrinsic knowledge representation method, and compare it to intrinsic knowledge in large language models for ontology learning. Our experiments show a trade-off between performance and semantic grounding for extrinsic knowledge compared to a fine-tuned model's intrinsic knowledge. We report our findings on the Large Language Models for Ontology Learning (LLMs4OL) 2024 challenge.

TSOTSALearning at LLMs4OL Tasks A and B : Combining Rules to Large Language Model for Ontology Learning

This paper presents our contribution to the Large Language Model For Ontology Learning (LLMs4OL) challenge hosted by ISWC conference. The challenge involves extracting and classifying various ontological components from multiple datasets. The organizers of the challenge provided us with the train set and the test set. Our goal consists of determining in which conditions foundation models such as BERT can be used for ontologies learning. To achieve this goal, we conducted a series of experiments on various datasets. Initially, GPT-4 was tested on the wordnet dataset, achieving an F1-score of 0.9264. Subsequently, we performed additional experiments on the same dataset using BERT. These experiments demonstrated that by combining BERT with rule-based methods, we achieved an F1-score of 0.9938, surpassing GPT-4 and securing the first place for term typing on the Wordnet dataset.

A Hybrid Approach to Ontology Construction for the Badini Kurdish Language

Semantic ontologies have been widely utilized as crucial tools within natural language processing, underpinning applications such as knowledge extraction, question answering, machine translation, text comprehension, information retrieval, and text summarization. While the Kurdish language, a low-resource language, has been the subject of some ontological research in other dialects, a semantic web ontology for the Badini dialect remains conspicuously absent. This paper addresses this gap by presenting a methodology for constructing and utilizing a semantic web ontology for the Badini dialect of the Kurdish language. A Badini annotated corpus (UOZBDN) was created and manually annotated with part-of-speech (POS) tags. Subsequently, an HMM-based POS tagger model was developed using the UOZBDN corpus and applied to annotate additional text for ontology extraction. Ontology extraction was performed by employing predefined rules to identify nouns and verbs from the model-annotated corpus and subsequently forming semantic predicates. Robust methodologies were adopted for ontology development, resulting in a high degree of precision. The POS tagging model attained an accuracy of 95.04% when applied to the UOZBDN corpus. Furthermore, a manual evaluation conducted by Badini Kurdish language experts yielded a 97.42% accuracy rate for the extracted ontology.

Fuzzy deep visual panic detection

Deep visual data analysis from social network has become an increasingly important area of research. In fact, this form of assessment makes it viable to recognize new information on social users which incorporates emotions. In order to recognize users’ emotions and other latent attributes, most of the existing approaches have used textual data and have obtained accurate results. However, little attention has been paid to visual data that have become increasingly popular in recent years.This work describes how to develop a conceptual representation model for social network analysis and social emotion analysis based on the integration of fuzzy logic and ontological modeling. The primary aim is to create an ontology that can extract new information about a social user’s mood, which can be classified as panic, no-panic, or neutral. Fuzzy logic is necessary to deal with subjective data, as users share imprecise and vague data on their social profiles. Fuzzy logic has been observed as a successful method to capture the expression of emotions due to the fuzzy nature of emotions and the ambiguous definitions of emotion words. The proposed work investigate the role of fuzzy logic in social network analysis. This study simulate a fuzzy deep system integrated with ontology for classifying social visual data (shared images) into panic, no-panic or neutral classes in order to determine the social users’ stress intensity. The Social distancing and the huge amount of shared data in Tunisia were calculated to demonstrate this classification. The experiments performed in this paper aim to create not only a novel annotated visual database named visual panic database, but also a new semantic model for modeling users’ profiles, in social network, based on the combination between ontology and deep learning techniques. In this context, we will work in the future on the combination between a user’s visual and textual data in order to improve the performance of this recognition. The proposed fuzzy system reflected the viral proliferation in stressed users and achieved an accuracy of 87%.

Review on integration of ontology and deep learning in cultural heritage image retrieval

Image retrieval methods are currently developing towards big data processing. The literature review is focused on image big data extraction with cultural heritage domain as training and testing datasets. The development of image retrieval process starts from content-based using machine algorithms, deep learning to ontology-based. Image recognition research with cultural heritage domain is conducted because of the importance of preserving and appreciating cultural heritage, in this case, cultural heritage images such as Indonesian Batik are discussed. Batik motif images are Indonesian cultural heritage that has thousands of motifs that are grouped into many classes with a non-linear hyperplane. The problem is focused on processing big data that has many classes. Currently research is evolving into knowledge-based image retrieval using ontologies due to semantic gap constraints. The results of this literature study can be the basis for developing research on the application of appropriate deep learning algorithms so as to utilize the hierarchy of classes and subclasses of image ontologies with cultural heritage domains.

Quantum ontology de-naturalized: What we can't learn from quantum mechanics

Philosophers of science commonly connect ontology and science, stating that these disciplines maintain a two-way relationship: on the one hand, we can extract ontology from scientific theories; on the other hand, ontology provides the realistic content of our scientific theories. In this article, we will critically examine the process of naturalizing ontology, i.e., confining the work of ontologists merely to the task of pointing out which entities certain theories commit themselves to. We will use non-relativistic quantum mechanics as a case study. We begin by distinguishing two roles for ontology: the first would be characterized by cataloging existing entities according to quantum mechanics; the second would be characterized by establishing more general ontological categories in which existing entities must be classified. We argue that only the first step is available for a naturalistic approach; the second step not being open for determination or anchoring in science. Finally, we also argue that metaphysics is still a step beyond ontology, not contained in either of the two tasks of ontology, being thus even farther from science.

Precise Molecular Subtyping Reveals Heterogeneity of Lung Adenocarcinoma Based on DNA Methylation

Background: Due to the high heterogeneity of lung adenocarcinoma (LUAD), which restricts the effectiveness of therapy, precise molecular subgrouping of LUAD is of great significance. Clinical research has demonstrated the significant potential of DNA methylation as a classification indicator for human malignancies. Methods: WGML framework (which was developed based on weighted gene correlation network analysis (WGCNA), Gene Ontology (GO), and machine learning) was developed to precisely subgroup molecular subtypes of LUAD. This framework included two parts: the WG algorithm and the machine learning part. The WG algorithm part was an original algorithm used to obtain a crucial module, which was characterized by weighted correlation network analysis, functional annotation, and mathematical algorithms. The machine learning part utilized the Boruta algorithm, random forest algorithm, and Gradient Boosting Regression Tree algorithm to select feature genes. Then, based on the results of the WGML framework, subtypes were computed by the hierarchical clustering algorithm. A series of analyses, including dimensionality reduction methods, survival analysis, clinical stage analysis, immune infiltration analysis, tumor environment analysis, immune checkpoints analysis, TIDE analysis, CYT analysis, somatic mutation analysis, and drug sensitivity analysis, were utilized to demonstrate the effectiveness of subgrouping. GEO datasets were used to externally validate the results. Meanwhile, another subgrouping method of LUAD from another study was employed to compare with the WGML framework. Result: By importing DNA methylation data into the WGML framework, nine genes were obtained to further subgroup LUAD. Three subtypes, the Carcinogenesis subtype, Immune-infiltration subtype, and Chemoresistance subtype, were identified. The dimensionality reduction method exhibited great distinctness between subtypes. A series of analyses were employed to exhibit the difference among the three subtypes and to demonstrate the accuracy of the definition of subtypes. Besides, the WGML framework was compared with a LUAD subgrouping method from another research, which demonstrated that WGML had better efficiency for subgrouping LUAD. Conclusion: This study provides a novel LUAD subgrouping framework named WGML for the accurate subgrouping of lung adenocarcinoma. result: By importing DNA methylation data into WGML framework, nine genes were obtained to further subgroup LUAD. Three subtypes, Carcinogenesis subtype, Immune-infiltration subtype, and Chemoresistance subtype were identified. Dimensionality reduction method exhibited great distinctness between subtypes. A series analyses were employed to exhibit the difference of three subtypes and to demonstrate accuracy of definition of subtypes. Besides, WGML framework was compared with a LUAD subgrouping method from another research, which demonstrated WGML had better efficiency for subgrouping LUAD.

Interpreting the structure and results of a data warehouse model using ontology and machine learning techniques

Interpreting the structure and results of a data warehouse model using ontology and machine learning techniques

From specialized knowledge frames to linguistically based ontologies

This paper explains conceptual modeling within the framework of Frame-Based Terminology (Faber, 2012; 2015; 2022), as applied to EcoLexicon (ecolexicon.ugr.es), a specialized knowledge base on the environment (León-Araúz, Reimerink &, Faber, 2019; Faber & León-Araúz, 2021). It describes how a frame-based terminological resource is currently being restructured and reengineered as an initial step towards its formalization and subsequent transformation into an ontology. It also explains how the information in EcoLexicon can be integrated in environmental ontologies such as ENVO (Buttigieg, Morrison, Smith, Mungall & Lewis, 2013; Buttigieg, Pafilis, Lewis, Schildhauer, Walls & Mungall, 2016), particularly at the bottom tiers of the Ontology Learning Layer Cake (Cimiano, 2006; Cimiano, Maedche, Staab & Volker, 2009). The assumption is that frames, as a conceptual modeling tool, and information extracted from corpora can be used to represent the conceptual structure of a specialized domain.

APROXIMAÇÕES ENTRE LEV VIGOTSKI E GŸORGŸ LUKÁCS: POR UMA ONTOLOGIA SOCIAL DA APRENDIZAGEM

This article aims to discuss the theoretical approaches between Lukács and Vygotski with regard to education, presenting some points that can demonstrate that there is a social ontology of human learning, that is, there is a specific way for human beings to learn that is not natural, but social and which implies the development of some assumptions (categories prior to learning itself), such as Work, Consciousness, Communication and Language. Based on Historical-dialectic Materialism, it analyzes the texts of Lukács and Vygotski, considering as main sources chapter II of the work “Towards an ontology of social being II” in the lukacsian work, called “Reproduction” and the Vygotskian work: “The construction of thought and language”, in your chapter 4 “The genetic roots of thought and language”. It presents a study regarding an ontology of education, whose principles are anchored in a historical and sociogenetic conception of the development of human capabilities. It understands the relationship between the development of Higher Psychological Functions and the education category as aspects of the ontological leap towards the social being, mediated by work activity. Finally, it recognizes the proximity between Vigostkian and Lukacsian theories based on work as a central category and mediator of the transformation of humanity's biological being into a social being. Other categories would emerge from work, such as language, education and consciousness, as well as presumably learning.

The Semantics of COVID-19 Web Data: Ontology Learning and Population

Background: The acquisition and exchange of meaningful, integrated, and accurate information are at the forefront of the combat against COVID-19; still, there are many countries whose health systems are disrupted. Moreover, no one is adequately equipped for COVID-19 contingencies. Many organizations have established static information systems to manage the information. This fact presents numerous issues, including delays, inconsistencies, and inaccuracies in COVID-19 information collected for pandemic control and monitoring. Objective: This paper presents a semantic representation of COVID-19 data, a domain ontology to facilitate measurement, clarification, linking, and sharing. We automatically generate a computer- intelligible knowledge base from COVID-19 case information, which contains machineunderstandable information. Furthermore, we have anticipated an ontology population algorithm from tabular data that delivers interoperable, consistent, and accurate content with COVID-19 information. Methods: We utilized the tabula package to extract the tables from PDF files and user NLP libraries to sort and rearrange tables. The proposed algorithm was then applied to all instances to automatically add to the input ontology using the Owlready Python module. Moreover, to evaluate the performance, SPARQL queries were used to retrieve answers to competency questions. Results: When there is an equivalence relationship, the suggested algorithm consistently finds the right alignments and performs at its best or very close to it in terms of precision. Moreover, a demonstration of algorithm performance and a case study on COVID-19 data to information management and visualization of the populated data are also presented. Conclusion: This paper presents an ontology learning/matching tool for ontology and populating instances automatically to ontology by emphasizing the importance of a unit's distinguishing features by unit matching.