Today, we are gathering more and more new astrophysics knowledge, but we are using obsolete ways of representing and processing it. This position paper discusses potential ways of constructing an astronomical KG, semantically annotating and reasoning on such data, using neuro-symbolic methods. AI is already revolutionizing knowledge acquisition and management, allowing computers to understand and process resources that are otherwise consumable only by humans. A common requirement for such technologies to operate successfully is a machine-readable conceptual modelling of the domain of interest. Having such a model enables, in turn, reasoning on the data to infer new knowledge. Typically, domain-specific models (e.g., SNOMED in healthcare, GeneOntology in genomics) are stored in knowledge graphs (KGs), formatted as ontologies.
2. Managing Astrophysics Data with CosmOntology
In recent years, there have been a plethora of works that try to exploit the potential of tables
available on the Web for a multitude of applications, ranging from knowledge base augmentation
CEUR
Workshop
Proceedings
to question answering, schema linking, and data integration [
In more detail, for the generation and curation of CosmOntology, a human-in-the-loop
approach is probably more promising, in which a board of experts will handle requests to update
the ontology. Such requests will be made both by algorithms and by humans. A set of many
shallow ontologies can be generated initially from structured data, which will be later enriched
by combining the many, shallow ontologies into one [
A recent work [
Applications. A unified KG modeling of astronomical concepts will allow a number of
AI tools and methods to become available. For example, accessing astrophysics data even
with natural language interfaces [
Some of the reasoning tasks that could be enabled after such a KG construction are: managing
inconsistencies [
Provenance and explainability are also major issues for astronomers, determining the credibility of the underlying KG. Data and workflow provenance information can be used for explainability, answering questions like “WHY am I (not) seeing this result?” and “HOW was this data acquired?” (e.g., with which instruments, under which conditions), respectively.
This paper discusses some possible steps for the construction of a KG that captures our knowledge of the cosmos, mostly relying on tabular data found in astrophysics publications and catalogues. Such a KG will boost research in astrophysics, widen the public knowledge in astronomy, and pose new challenges that can further improve research in computer science. For example, the resources produced by this efort could be used to improve deep-learning tools that mine such data, or semantically annotate them (e.g., systems participating in SemTab). The processes followed in this endeavor should be thoroughly documented, with the ultimate goal to create a generalizable, open-access methodology that gets adopted by other disciplines. 2005. [21] S. Schlobach, R. Cornet, Non-standard reasoning services for the debugging of description logic terminologies, in: IJCAI, 2003. [22] B. C. Grau, I. Horrocks, Y. Kazakov, U. Sattler, Modular reuse of ontologies: Theory and practice, J. Artif. Intell. Res. 31 (2008) 273–318. [23] A. Bonifati, S. Dumbrava, H. Kondylakis, Graph summarization, CoRR abs/2004.14794 (2020).