Page 31 - Sirius_Annual_Report_2021

designed user interface, grown out of a long experience
in data integration projects is completely dedicated to
the technology of Virtual Knowledge Graphs. Figure 2 is a screenshot of Ontopic Studio for editing Mapping. There are many innovative new functionalities that support the data architect for the whole lifecycle of the Knowledge Graph. Ontopic Studio internally relies on Ontop, the most advanced Open-Source VKG engine.
New Reasoning Techniques
CQ answering [5][6][7]. We continued our work on opti- mising CQ answering over unrestricted OWL 2 ontologies. This effort resulted in the development of two tools:
• RSAComb, an optimised implementation of a combined approach technique for conjunctive query answering over RSA ontologies.
• The next iteration in the development of PAGOdA, now integrating RSAComb to improve the computation of query bounds.
We extended our research on ontology approximations; we proposed a novel technique to compute an RSA restriction of an unrestricted ontology that maintains completeness w.r.t. CQ answering. The most recent evaluation of the system provides a fair and extensive analysis of the tools’ performance and capabilities.
KG construction, curation, etc. [8][9][10][11]. We mainly worked on ontology alignment by combining modern deep learning techniques such as pre-trained language models and distant supervision with traditional ontology alignment systems such as LogMap. Specifically, we developed LogMap-ML which is an extension for LogMap with higher results especially w.r.t. Recall. It uses LogMap anchor map- pings filtered by disjointness-based rules to train a mapping prediction model which uses two classes’ embeddings or their paths’ embeddings as the input. We also developed
a new system named BERTMap which totally gives up the lexical mapping part of LogMap but uses the pre-trained language model BERT to fully utilize the textual information. BERTMap can achieve state-of-the-art performance on the tasks of the OAEI LargeBio track.
We also continued the works of knowledge graph correction and OWL ontology embeddings. Our previous Web conference paper on erroneous fact correction is extended with the case of correcting mapping assertions, and the extension is
accepted by Semantic Web Journal. Our Word2Vec-based ontology embedding method OWL2Vec* is finally published in Machine Learning Journal.
Use Cases
Our approach has been successfully deployed in a few novel use cases. We have developed the SIRIUS OBDA subsurface pilot, and applied it to the South Tyrolean OpenDataHub, LinkedGeoData, Dow Jones, and Festo.
The SIRIUS OBDA subsurface pilot project is addressing these shortcomings and aim to significantly broaden the applicability of the approach for use in subsurface projects. In the past, the Optique project was focused on developing OBDA on a relational database at Equinor, which is no longer an active database. Further, that database contains proprietary data, and access to that is now restricted. This resulted in a significant setback for researchers to continue further extermination on extending OBDA capabilities. In 2021 we established a large in-house relational database from the publicly available G&G datasets (mainly by processing the Volve dataset https:/data.equinor.com & NPD FactPages https:/factpages.npd.no). This database is now being utilized in various other internal and external research projects where G&G data is being used for experimentation, e.g., DigiWell at USN is also using this database for research purposes. We have adapted mapping and Ontology from Optique, and carried out Integration & testing. SIRUS OBDA Subsurface V1.0 was demonstrated at the SIRIUS General Assembly in November 2021.
South Tyrol OpenDataHub Knowledge Graph (https:/ sparql.opendatahub.bz.it/) is a joint project between NOI Techpark and Ontopic for publishing South Tyrolean tourism data as a Knowledge Graph. LinkedGeoData (http:/linked geodata.org/) is an effort to add a spatial dimension to the Semantic Web. LinkedGeoData uses the information collected by the OpenStreetMap project and makes it available as an RDF knowledge base according to the Linked Data principles [3]. The Festo case study (https:/uploads-ssl.webflow.com/ 5ed7f18d11a068aa460ce2e9/5f5252796dd12f613510c1eb _Festo%20Case%20Study.pdf) shows how Festo was able to completely transform the related internal data processes to reduce the time to provide satisfactory specifications from hours to seconds using RDFox. OST’s partner Derivo integrated RDFox in Festo’s Semantic Platform. Finally,
for Dow Jones, RDFox enhances the scope and capability of various products, from The Wall Street Journal to competitor listings in the S&P 500 index (https:/www.oxfordsemantic. tech/blog/dow-jones-enhances-product-line-with-semantic- innovation).
SIRIUS ANNUAL REPORT 2021 | 31