Machine Learning for Ontology Alignment

The main objective of this project is to extend the state of the art in ontology alignment with machine learning techniques to enhance the alignment process and adaptability of current ontology alignment systems to cope with new domains and tasks.

Project Background and Scientific Basis

Ontologies are extensively used in biology and medicine. Ontologies such as SNOMED CT, the National Cancer Institute Thesaurus (NCI), and the Foundational Model of Anatomy (FMA) are gradually superseding existing medical classifications and are becoming core platforms for accessing, gathering and sharing bio-medical knowledge and data. These reference biomedical ontologies, however, are being developed independently by different groups of experts and, as a result, they use different entity naming schemes and modelling conventions. As a consequence, to integrate and migrate data among applications, it is crucial to first establish correspondences (or
mappings) between the vocabularies of their respective ontologies.

In the last ten years, the Semantic Web and biomedical research communities have extensively investigated the problem of automatically computing mappings between independently developed ontologies, usually referred to as the ontology matching problem (see [1] for a comprehensive and up-to-date survey). The growing number of available techniques and increasingly mature tools, together with substantial human curation effort and complex auditing protocols, has made the generation of mappings between real-world ontologies possible.

The use of machine learning to enhance the alignment process is not new, however state of the art systems like SILK, LIMES or YAM++ [3-5] do not exploit available alignments (e.g., precomputed candidate alignments or in public repositories) to learn or predict new links nor involve the ontological information within the learning process, which limits the versatility and adaptability of these systems, especially when dealing with complex tasks. In this MSc project we aim performing a preliminary study of the suitability of machine learning techniques to predict new alignments using available alignment repositories (e.g., [6-7]).

Supervision

The thesis will be jointly supervised by Dr. Ernesto Jimenez-Ruiz and Prof. Martin Giese from the Logic and Intelligent Data (LogID) group, based in the Department of Informatics.

The LogID group is also actively contributing to the Ontology Matching community and (co)organises the annual Ontology Alignment Evaluation Initiative (OAEI). The OAEI [2] is an annual campaign for the systematic evaluation of Ontology Alignment systems.

References

[1] Pavel Shvaiko, Jérôme Euzenat: Ontology Matching: State of the Art and Future Challenges. IEEE Trans. Knowl. Data Eng. 25(1): 158-176 (2013)

[2] Manel Achichi, et al.: Results of the Ontology Alignment Evaluation Initiative 2016. OM@ISWC 2016: 73-129

[3] A. N. Ngomo et al. LIMES - A Time-Efficient Approach for Large-Scale Link Discovery on the Web of Data. In IJCAI, 2011

[4] D. Ngo et al. YAM++ : A multi-strategy based approach for ontology matching task. In EKAW, 2012.

[5] J. Volz et al. Discovering and maintaining links on the web of data. In ISWC, 2009.

[6] EBI’s Ontology Lookup Service and related tools. http://www.ebi.ac.uk/ols/index

[7] N. F. Noy et al. BioPortal: ontologies and integrated data resources at the click of a mouse. Nucleic Acids Res, 2009

 

Publisert 11. aug. 2017 11:46 - Sist endret 28. sep. 2017 22:02

Omfang (studiepoeng)

60