OBARIS: Ontology-Based Artificial Intelligence in Environmental Sector

The OBARIS project aims to advance the status quo in auditable semantic artificial intelligence systems (SAIS) by investigating both conceptual aspects of these systems as well as the development of a technology stack that transfers these system types into concrete settings.

OBARIS covers a range of innovative aspects:

The first aspect is to improve the understanding of the different types of semantic AI systems (SAIS) on a conceptual level. We further refined the concept of SAIS into Semantic Web and Machine Learning System (SWeMLS). One important outcome is a systematically derived taxonomic characterisation of SWeML systems and their patterns based on a large-scale Systematic Mapping Study.

The second aspect was the development of a lifecycle model, which helps identify and guide the development of concrete SWeMLS implementations. Based on this lifecycle model, we devised a concrete technology stack for developing SWeMLS in the OBARIS showcases. To this end, we developed a flexible pipeline architecture that adopted the widely used NLP Interchange Format (NIF). The pipeline communicates with a range of services which ensure that it can be reused and extended.

The third aspect was related to the design and development of an auditability framework for SWeMLS. To this end, we have developed a semantic-based method and models which allow automated capturing of audit traces and machine-actionable description of SWeMLS. We reuse and extend the PROV-O and P-Plan ontologies to represent SWeMLS workflow and to generate corresponding endpoints for system trace acquisition.

These innovative aspects highlight the challenges that we addressed within the scope of the OBARIS project:

• the lack of understanding of the characteristics of SWeMLS
• the need for a generic and practical framework to support the auditability of SWeMLS
• the limited application of SWeMLS in the environmental domain.

We presented the software that we have developed to our project partners in two specific use cases in the environmental sector and evaluated them together. The first showcase deals with machine learning-supported analysis of legal permit information. The second showcase is concerned with the FAIRification of nutrient input and flows of river catchments in Austria, i.e. the collection and integration of heterogeneous measurement data and metadata from different sources in order to increase data transparency.

These innovative aspects and their usage within the specific showcases highlight the contributions made by OBARIS to the SDGs, in particular to Goal 9 "Industry, innovation and infrastructure". First of all, our results contribute to understanding the new wave of semantic AI systems and the auditability aspects of these systems. Secondly, we have developed two proofs-of-concept based on our research results that demonstrate the applicability and further development of these technologies.

Our plan is to develop SWeMLS design patterns further in future and explore their impact on auditability and other system characteristics. We will also continue to develop our software so as to add further functionalities and increase user-friendliness.

Consortium leader

• TU Wien, Data Science research unit (194-04)

Additional consortium partners

• Semantic Web Company (SWC)
• Environment Agency Austria (UBA)