Ontology networks (ON) have been arising as robust frameworks for organizing, linking, and managing vast amounts of information across diferent domains. However, the development of an ontology network requires the establishment of accurate and robust intra-domain and inter-domain relationships. Intradomain relationships refer to connections between concepts from diferent ontologies from the same domain, while inter-domain relationships involve connections across diferent domains. We believe that ontology matching can play a crucial role when creating these relationships, as ontology matching serves as a pivotal mechanism for aligning and integrating ontologies to enable meaningful interactions between various domains. Through ontology matching, disparate ontologies can be harmonized and linked, thereby facilitating efective knowledge sharing and interoperability. In this paper, we focus on constructing intra-domain links using a voting-based matching approach. In particular, these links are generated using a voting system using the alignments generated by a number of well-known matching systems.
CEUR ceur-ws.org
In the rapidly evolving landscape of knowledge representation and data integration, ontology
networks (ON) have emerged as powerful frameworks for organizing, linking, and managing
vast amounts of information. An ontology network or a network of ontologies is a collection
of individual ontologies interconnected through various relationships, including alignment,
modularization, and dependency relationships [
A second current use case for ontology development can be observed within diferent NFDI (National Research Data Infrastructure)1 consortia in Germany. The NFDI aims to create a common, sustainable infrastructure for research data management in Germany. It is composed of around 30 discipline-specific consortia. In the following, we focus on NFDI-MatWerk, the consortium for material sciences and engineering. Similar issues exist in many of the other domains, though. NFDI-MatWerk, explicitly addresses the significant role of ontology matching2. Ontology matching is the process of identifying and aligning concepts and relationships between two or more ontologies [4]. One crucial use case involves establishing connections between domain-specific ontologies and application ontologies with more general or foundational ontologies. This process plays a vital role in enabling interoperability between various sub-domains and is essential for linking diferent sub-domains within the NFDI-MatWerk framework. Furthermore, ontology matching serves as a prerequisite for creating a well-connected “network” of ontologies across diferent NFDI consortia. The successful alignment and integration of ontologies from various consortia enable the establishment of strong intra-domain and inter-domain relationships, enhancing the overall coherence and comprehensiveness of the NFDI infrastructure. This is essential, as many scientific questions require integration of data in the responsibility of several diferent NFDI consortia.
The primary aim of this study is to explore the inherent potential of ontology matching techniques in developing ontology networks. Our proposed ON is based on a three-level architecture comprising foundational, core, and domain ontologies. We curate ontologies from various domains, utilize multiple well-known matching systems for generating alignments, and adopt a voting-based matching approach to create intra-domain links.
Ontology networks establish connections between multiple ontologies, utilizing various relationships such as mapping, modularization, and versioning [5]. Recent literature showcases numerous studies implementing diverse methodologies to develop ontology networks across
2https://nfdi-matwerk.de/infrastructure-use-cases/iuc12-alignment-of-application-and-higher-level-ontologies
multiple domains [
A comprehensive Ontology Network for Diabetes Mellitus in Mexico was developed, incorporating six distinct domains and introducing new classes by leveraging ontological and non-ontological resources, resulting in 1367 classes, 20 object properties, 63 data properties, and 4268 individuals sourced from seven diferent ontologies [ 9]. The methodology comprises several steps, commencing with defining network elements and participating domains, along with their scope, and performing ontological engineering tasks such as design, reuse, population, and evaluation, with a particular emphasis on the concept of ON. In their approach, the ontology integration process involves several steps: determining the base ontology for enrichment, identifying elements with similar names, conducting semantic verification, analyzing the representation of common elements (classes, instances, and properties), determining the final structure, and evaluating the resulting ontology. In analyzing the representation of common elements, they found that the best option is to convert the lowest classification levels to instances and to instance them from the immediate superior class. Our approach emphasizes on voting-based matching for intra-domain links, while their work explores ontology integration and new class introductions in specific domains.
SEON, a software engineering ontology network (ON) [
Ontology matching techniques play a crucial role in ontology network development. OM provides a solution to the issue of semantic heterogeneity by identifying correspondences between semantically related entities across ontologies [4]. A number of ontology matching systems have been developed in the past, demonstrating efective performance in real-world scenarios [11, 12, 13, 14, 15]. These systems typically use a combination of automated and manual techniques to identify correspondences between ontologies. Silva et al. [16] employ ontology matching (OM) techniques to create a network of 28 integrated ontologies, forming a knowledge graph for Explainable AI in personalized oncology. Their approach introduces a novel holistic ontology alignment strategy, using AML [17], which clusters ontologies based on semantic overlap measured by fast matching techniques with a high confidence level, followed by applying more sophisticated matching techniques within each cluster. In contrast, our approach concentrates on developing ontology networks by establishing intra-domain and inter-domain links through a voting-based matching approach, utilizing multiple well-known 3https://github.com/spice-h2020/SON, https://bimerr.iot.linkeddata.es/, https://github.com/rapw3k/glosis matching systems, including AML, for generating required alignments. They also provide challenges of integrating 28 ontologies to form a KG at three levels: biomedical ontology matching, holistic ontology matching, and holistic ontology integration, in addition to quality, coverage, and scalability. High-quality mappings are vital, considering the minimal human involvement in large-scale tasks. Suficient coverage of all domains is imperative to ensure comprehensive data description, supporting AI model training and explanations. Lastly, the scalability of the system must be considered due to the extensive processing of a large number of classes and properties from various ontologies. Balancing quality, coverage, and scalability is essential in building a high-quality network of biomedical ontologies.
Santos et al. [18] propose a diferent approach by utilizing random walks and frequent item sets algorithms to mine data from the networks, identifying relevant candidate entities and then pruning the networks using an algebraic method to remove identical entities, and subsequently reintegrating the relevant nodes to preserve essential correspondences. Foundational ontologies, a part of ONs, also play a crucial role in ontology matching by ofering a well-founded reference model that can be shared across diverse domains. In the paper [19], the authors provide an overarching perspective on the various tasks involved in ontology matching, taking into account the incorporation of foundational ontologies. In this paper, we do not currently employ foundational ontologies for matching, but it is a potential avenue for future research. Another paper [20] addresses the problem of adding new information in networks of ontologies based on belief revision using the semantics of networks of ontologies.
A recent work focuses on developing a framework for classifying ontologies that provides a homogeneous environment for ontology network development [21, 22]. It classifies ontologies into reference ontologies, operational ontologies, operational ontologies with a previous reference ontology, and well-grounded ontology. Compared to our proposed ON, we propose a three-layered network, including foundational, core and domain ontologies. The framework investigates diferent kinds of relationships between ontologies, such as groundover, drive, reuse, and subontology relations [21, 22]. It is very important to identify these kind of relationships between ontologies in the ON, but we need first to determine which concepts from diferent ontologies participate in these diferent kinds of relationships.
To construct intra-domain and inter-domain connections across given sets of ontologies, we develop a voting system that makes use of at least three well-known matching systems. The general architecture of the voting system has a number of components as shown in Fig. 1, including data ontology curating, matching, voting and validating. In the following, we are going to describe these components. Throughout the description we will use the creation of an ontology network for provenance information as a running example. The motivation for establishing this ontology network is elaborated in our previous work [23].
Our study aims to develop an ontology network for seamlessly integrating data from diverse scientific studies within an interdisciplinary research project. To achieve this, we integrate existing ontologies from various domains rather than building them from scratch. Our proposed architecture adopts a three-layer view, comprising foundational, core, and domain ontologies, with the aim of integrating diverse aspects of provenance of scientific study, which includes both computational and non-computational aspects across multiple applications [3], thus promoting reproducibility support. We identified existing ontologies in various domains through a systematic literature review based on our research requirements [3, 23]. Utilizing Google Scholar, we specifically focused on the areas of Scientific Experiments , Machine Learning, Microscopy, Computational, Scientific Workflows , and Provenance to compile a comprehensive list of ontologies in addition to foundational ontologies. In this process, we added additional ontologies to our existing collection4.
We found nine ontologies in Provenance, seven in Scientific Experiments , three in Microscopy, nine in Computational, six ontologies (thirteen with sub ontologies) in the Machine Learning domain, and four foundational ontologies. We collected the year of creation and the last update of each ontology. Furthermore, we conducted checks to verify the current availability and accessibility status of each ontology mentioned in the collected publications. In the next step, we excluded certain ontologies due to their unavailability or being in a format diferent from the standard ontology format. The resulting list of ontologies from Scientific Experiments , Microscopy, Computational and Machine Learning domains5, along with their version, number of classes, and properties, is presented in Table 1.
The cumulative number of classes across these domains are 10,259, while the number of properties amounts to 1659. Frequently, ontologies undergo regular modifications and are not static. This evolution involves incorporating new domain knowledge, rectifying design errors, or adapting to revised requirements. The active development of certain ontologies is evident in their version or release updates, reflecting ongoing improvements and enhancements (e.g., SMART Protocols, CMPO, EDAM); others remain stable over time (e.g., EXPO, MEX). Additionally, it is worth noting that certain ontologies lack version information. Furthermore, the number of classes varies significantly, ranging from 1 (CSO) to 3473 (EDAM), and the
5In this paper, foundational ontologies and ontologies from Scientific Workflows and Provenance domains are not included in the matching step, but will be taken into consideration in future work. number of properties varies from 0 (CSO) to 265 (MEX). In the context of Microscopy, the ontologies exhibit class counts ranging from 17 to 1058, presenting challenges when performing the ontology matching task.
In the following, we first analyze our example domain to motivate the need for matching and then introduce our approach to it. 3.2.1. Motivation for matching Each ontology within the collected set of ontologies models a part of the domain it represents. However, frequently the same piece of information is represented in diferent ontologies. Sometimes, these representations are similar, sometimes they difer significantly. For example, as shown in Table 1, we collected six ontologies from the ML domain with a number of concepts ranging from 25 concepts (in the MLSchema ontology) to 697 (in the DMOP ontology). Investigating this set of ontologies, we found that it has a number of semantic heterogeneities that must be aligned before the construction of the ontology network. For example, the term “parameter” has been modeled in all six ML ontologies using diferent IRIs. It is defined using the ID “http://www.e-lico.eu/ontologies/dmo/DMOP/DMOP.owl#Parameter” in the BigOWL and DMOP ontologies, while diferent IDs have been used to represent it in the other four ontologies. Indeed, to establish the accurate links between ontologies from the ML domain, it is first required to identify these diferently modelled concepts as the same or diferent object. This is a perfect task for ontology matching.
The situation becomes worse when we attempt to construct links between similar concepts across diferent domains, as shown in Fig. 2. For example, the term “data” which is commonly used and needed in our use case, is defined using diferent terms across the five domains, “data”, “dataset”, “data item”, and “datatype”. Considering the term “dataset” which is modeled in four domains6. In SWO Computational ontology, there is a class called “data item”. Also, in EDAM ontology, it is “data”, as shown in Fig. 2. The term dataset is defined using IRI http://www.e-lico.eu/ontologies/dmo/DMOP/DMOP.owl#DataSetClass in the ML domain, and another IRI https://w3id.org/reproduceme#Dataset for the experimental domain. Another example is the term “model”. The class “model” has a diferent meaning in Machine Learning than the term in Microscopy. In Microscopy, the class model is a manufacturer specification for a device, while in ML, it is a file created from ML algorithms that has been trained to recognize 6We are expecting it to be also modeled in the computational domain if we consider more ontologies from the domain certain types of patterns based on previous experience or data.
The diversity of representations for ontology elements poses a significant challenge to ontology integration, encompassing issues like synonyms (representing the same thing with diferent names), homonyms (elements with the same name but diferent meanings), and varied correspondences between concepts in diferent ontologies. Without a well-defined strategy for integrating ontologies, inconsistency problems can arise that can alter the operation of ontologies. 3.2.2. Matching Approaches Once we collected relevant ontologies to construct the ontology network, the arising question is how to establish links between similar concepts from the same domain, intra-domain links and links between related concepts from ontologies across diferent domains inter-domain links. In this paper, we focus on the intra-domain connections. To this end, we make use of available matching systems. However, as a matching system exploits specific features from input ontologies to generate alignments, we propose to use not only one matching system but at least three (or any larger uneven number of) available and well-known matching systems to generate the required alignment: LogMap [24], AML [17], and OAPT [12, 14].
Matching tasks: There are two diferent options to generate matching tasks across ontologies from the same domain, namely pair-wise matching and holistic matching [16, 25]. In the pairwise matching scheme, every ontology pair is uploaded, and a matching algorithm is applied; in the holistic matching technique, a matching algorithm is applied to the whole set of ontologies from the same domain. It is unclear which matching outperforms the other, and as our proposed method is based on existing matching systems classified as pair-wise matching systems, we selected the pair-wise matching scheme. This results in ×(−21) matching tasks for ontologies in a domain.
Alignment candidate generation: A first step to construct the intra-domain links is to determine similar entities from diferent ontologies from the same domain. To this end, we generate match candidates using at least three ontology matching systems. We make use of logMap [24] and AML [17] as they are performing the best for OAEI (Ontology Alignment Evaluation Initiative)7 during the last ten years. Furthermore, we use our tool OAPT [12], the light version of LogMap [15] and PoMap++ matching system that participated in OAEI 2020 [13]. We run a matching system using one matching task according to the pair-wise scheme we consider. We save the matching result for each task for further investigation. Fig. 3 shows a match result example between two concepts from the REPRODUCE-ME ontology and the CMPO ontology.
This example is taken from the Microscopy domain. The figure shows that only four out of the ifve systems used identify the mentioned correspondence at all. Furthermore, matching systems that are able to discover the same correspondences, assign diferent similarity values. For example, the two concepts in Fig. 3 are "http://purl.obolibrary.org/obo/PATO_0000140"/ with a position as a label and "http://www.w3.org/ns/prov#Location"/ with Location as a label. Both POMap++ and LogMapLite are able to discover this correspondence with a similarity value of 1.0, while the main LogMap system can achieve a similarity value of 0.51. The interesting ifnding from Fig. 3 is that LogMapLite is not able to determine the correspondence between the mentioned two entities. The example shown in Fig. 3 highlights the importance of using several matchers to increase the chance of finding all relevant correspondences, but also the need to find a way to combine results from diferent matchers. In our approach, this is achieved via voting.
The consensus alignments will be generated by making use of a voting algorithm utilizing alignments generated by diferent matching systems[ 26, 27]. A vote considers the number of times a mapping appears in the set of generated alignments by diferent matching systems. For example, the consensus of vote 3 means that a mapping is proposed by at least three matching systems. In general, the more the votes, the smaller the size of the consensus alignment.
Tables 2 and 3 show the sizes of the consensus alignments for the ML and Experiment matching tasks, respectively, as well as each systems family mappings ratio contribution.
After generating diferent sets of alignments, we carried out a validation process based on our knowledge and experience. Two authors of the paper have manually looked at diferent alignments. We plan to extend the validation process by involving more domain experts from System Total mapping AML LogMap LogMap-Lt OAPT POMap each domain to cross-validate our initial validation.
The resources related to the paper are published on GitHub ‘https://github.com/fusion-jena/ ReproduceMeON’. This includes the ontology dataset, the methodology used to extract the ontology dataset, matching system results, and the output of the voting algorithm.
In conclusion, our paper presents a novel approach for developing ontology networks through the integration of ontology matching techniques, with a specific focus on establishing accurate and robust intra-domain links. By adopting a voting-based matching approach and leveraging multiple well-known matching systems. Even we achieved good progress to establish intradomain links of ON, however, we plan to extend the validation process to involve more domain experts. Furthermore, once we settle intra-domain links, we will investigate how to use them to establish inter-domain links.
The authors thank the Carl Zeiss Foundation for the financial support of the project ’A Virtual Werkstatt for Digitization in the Sciences (K3)’ within the scope of the program line ’Breakthroughs: Exploring Intelligent Systems for Digitization - explore the basics, use applications’. A. Algergawy’work has been partially funded by the Deutsche Forschungsgemeinschaft (CRC AquaDiva, Project 218627073). [3] S. Samuel, B. König-Ries, End-to-end provenance representation for the understandability and reproducibility of scientific experiments using a semantic approach, J. Biomed. Semant. 13 (2022) 1. doi:10.1186/s13326- 021- 00253- 1. [4] J. Euzenat, P. Shvaiko, Ontology matching, volume 18, Springer, 2007. [5] A. Dıaz, R. Motz, E. Rohrer, Making ontology relationships explicit in a ontology network,
AMW 749 (2011). [6] C. Sila, O. Belo, V. Barros, Methodology for the development of an ontology network on the brazilian national system for the evaluation of higher education (ontosinaes), JISTEMJournal of Information Systems and Technology Management 15 (2018). [7] Y. Li, R. García-Castro, N. Mihindukulasooriya, J. O’Donnell, S. Vega-Sánchez, Enhancing energy management at district and building levels via an em-kpi ontology, Automation in Construction 99 (2019) 152–167. doi:https://doi.org/10.1016/j.autcon.2018.12. 010. [8] S. D. Costa, M. P. Barcellos, R. de Almeida Falbo, M. V. H. B. Castro, Towards an ontology network on human-computer interaction, in: Conceptual Modeling - 39th International Conference, ER 2020, Vienna, Austria, November 3-6, 2020, Proceedings, 2020, pp. 331–341. doi:10.1007/978- 3- 030- 62522- 1\_24. [9] C. Reyes-Peña, M. Tovar, M. Bravo, R. Motz, An ontology network for diabetes mellitus in mexico, Journal of Biomedical Semantics 12 (2021) 1–18. [10] O. Corcho, D. Chaves-Fraga, J. Toledo, J. Arenas-Guerrero, C. Badenes-Olmedo, M. Wang, H. Peng, N. Burrett, J. Mora, P. Zhang, A high-level ontology network for ict infrastructures, in: The Semantic Web–ISWC 2021: 20th International Semantic Web Conference, ISWC 2021, Virtual Event, October 24–28, 2021, Proceedings 20, Springer, 2021, pp. 446–462. [11] P. Shvaiko, J. Euzenat, Ontology matching: state of the art and future challenges, IEEE
Transactions on knowledge and data engineering 25 (2011) 158–176. [12] A. Algergawy, S. Babalou, M. J. Kargar, S. H. Davarpanah, Seecont: A new seeding-based clustering approach for ontology matching, in: Advances in Databases and Information Systems - 19th East European Conference, ADBIS 2015, Poitiers, France, September 8-11, 2015, Proceedings, Springer, 2015, pp. 245–258. [13] A. Laadhar, F. Ghozzi, I. Megdiche, F. Ravat, O. Teste, F. Gargouri, Pomap++ results for OAEI 2019: Fully automated machine learning approach for ontology matching, in: Proceedings of the 14th International Workshop on Ontology Matching co-located with the 18th International Semantic Web Conference (ISWC 2019), Auckland, New Zealand, October 26, 2019, 2019. [14] A. Algergawy, S. Babalou, F. Klan, B. König-Ries, Ontology modularization with OAPT, J.
Data Semant. 9 (2020) 53–83. [15] E. Jiménez-Ruiz, Logmap family participation in the OAEI 2022, in: Proceedings of the 17th International Workshop on Ontology Matching (OM 2022) co-located with the 21th International Semantic Web Conference (ISWC 2022), Hangzhou, China, held as a virtual conference, October 23, 2022, 2022. [16] M. C. Silva, D. Faria, C. Pesquita, Matching multiple ontologies to build a knowledge graph for personalized medicine, in: The Semantic Web, Springer International Publishing, Cham, 2022, pp. 461–477. [17] D. Faria, C. Pesquita, E. Santos, I. F. Cruz, F. M. Couto, Agreementmakerlight 2.0: Towards eficient large-scale ontology matching, in: Proceedings of the ISWC 2014 Posters & Demonstrations Track, ISWC 2014, Riva del Garda, Italy, October 21, 2014, volume 1272 of CEUR Workshop Proceedings, CEUR-WS.org, 2014, pp. 457–460. URL: http://ceur-ws.org/ Vol-1272/paper_151.pdf. [18] F. Santos, C. E. Mello, Matching network of ontologies: A random walk and frequent itemsets approach, IEEE Access 10 (2022) 44638–44659. doi:10.1109/ACCESS.2022.3164067. [19] C. Trojahn, R. Vieira, D. Schmidt, A. Pease, G. Guizzardi, Foundational ontologies meet ontology matching: A survey, Semantic Web 13 (2022) 685–704. [20] J. Euzenat, Revision in networks of ontologies, Artif. Intell. 228 (2015) 195–216. doi:10.
1016/j.artint.2015.07.007. [21] B. F. Martins, L. J. S. Gil, J. F. R. Román, J. I. Panach, O. Pastor, M. Hadad, B. Rochwerger, A framework for conceptual characterization of ontologies and its application in the cybersecurity domain, Softw. Syst. Model. 21 (2022) 1437–1464. doi: 10.1007/ s10270- 022- 01013- 0. [22] B. F. Martins, J. F. R. Román, O. Pastor, M. Hadad, Improving conceptual domain characterization in ontology networks, in: Research Challenges in Information Science: Information Science and the Connected World - 17th International Conference, RCIS 2023, Corfu, Greece, May 23-26, 2023, Proceedings, volume 476 of Lecture Notes in Business Information Processing, 2023, pp. 187–202. doi:10.1007/978- 3- 031- 33080- 3\_12. [23] S. Samuel, A. Algergawy, B. König-Ries, Towards an ontology network for the reproducibility of scientific studies, in: Proceedings of the Joint Ontology Workshops 2021 Episode VII:, Bolzano, Italy, September 11-18, 2021, volume 2969 of CEUR Workshop Proceedings, CEUR-WS.org, 2021. URL: https://ceur-ws.org/Vol-2969/paper64-OntoCom.pdf. [24] E. Jiménez-Ruiz, B. C. Grau, Logmap: Logic-based and scalable ontology matching, in: The Semantic Web - ISWC 2011 - 10th International Semantic Web Conference, Bonn, Germany, October 23-27, 2011, Proceedings, Part I, 2011, pp. 273–288. [25] E. Rahm, The case for holistic data integration, in: Advances in Databases and Information Systems: 20th East European Conference, ADBIS 2016, Prague, Czech Republic, August 28-31, 2016, Proceedings 20, Springer, 2016, pp. 11–27. [26] E. Jiménez-Ruiz, B. C. Grau, I. Horrocks, Is my ontology matching system similar to yours?, in: Proceedings of the 8th International Workshop on Ontology Matching co-located with the 12th International Semantic Web Conference (ISWC 2013), 2013, pp. 229–230. [27] N. Karam, A. Khiat, A. Algergawy, M. Sattler, C. Weiland, M. Schmidt, Matching biodiversity and ecology ontologies: challenges and evaluation results, Knowl. Eng. Rev. 35 (2020) e9. doi:10.1017/S0269888920000132.