Materials science experiments involve complex data that are often very heterogeneous and challenging to reproduce. Challenges with materials science data were observed, for example, in a previous study on harnessing lightweight design potentials via the Materials Data Space for which the data from materials sciences engineering experiments were generated using linked open data principles, e.g., Resource Description Framework (RDF) as the standard model for data interchange on the Web. However, detailed knowledge of formulating questions in the query language SPARQL is necessary to query the data. It was noticed that domain experts in Materials Science lack knowledge of querying the data using SPARQL queries. With this work, we aim to develop NaturalMSEQueries an approach for the material science domain expert where instead of SPARQL queries, the user can develop expressions in natural language, e.g., English, to query the data. This will significantly improve the usability of Semantic Web approaches in materials science and lower the adoption threshold of the methods for the domain experts. We plan to evaluate our approach, with varying amounts of data, from different sources. Furthermore, we want to compare with synthetic data to assess the quality of the implementation of our approach.
1. www.bam.de
NATURALMSEQUERIES
A NATURAL WAY TO QUERY MATERIALS SCIENCE ENGINEERING DATA
EXPERIMENTS
Andre Valdestilhas, Thomas Hanke, Soudeh Javamasoudian, Ghezal
Ahmad Jan Zia, Horst Fellenberg, Thilo Muth
5. Data → Knowledge – FAIR- and Open-Data
5-15
I :
R :
A :
F :
REUSABLE
• Value creation: Creation of new knowledge
with fewer attempts or re-evaluation
FINDABLE
• What data exists?
• How & where do I find the measured
values?
ACCESSIBLE
• Is raw data & metadata accessible?
• -> Quality / value of data
• Restrictions? (Software, formats…)
INTEROPERABLE
• Usability beyond the originator:
-> Input & query (internal and external)
6. How MSE domain expert obtain the Knowledge
Graph
6-15
Input: Raw data from
Experiment and
Simulation
Mapping-Method Output: Knowledge
Graph
Metadaten and
Processed data
8. 8-15
Research questions
RQ1. How to query semantic MSE data
easier than using SPARQL queries?
RQ2. What is the best way to organize
Material Sciences Methods data?
RQ3. How much will the framework help
the Materials Science Engineering domain?
9. Methodology (our approach)
9-15
Sparklis citation: Ferré, Sébastien. ‘Sparklis: An Expressive Query Builder for SPARQL Endpoints with Guidance in Natural
Language’. Semantic Web 8(3) : 405-418. IOS Press, 2017
12. Natural Language to SPARQL query and graph
visualization of the query results
12-15
13. Evaluation: Domain experts creating
knowledge – Ontology, querying data
13-15
• Link data across institutions (BAM – Fraunhofer – KIT, etc)
• Exploit heterogenous materials data
14. Approach – Research Questions
14-15
RQ1. How to query semantic
MSE data easier than using
SPARQL queries? (NLP)
RQ2. What is the best way to
organize Material Sciences
Methods data?
(RDF Knowledge Graph)
RQ3. How much will the
framework help the Materials
Science Engineering domain?
(Evaluation/Usability)
15. Conclusion
1 Potentialfor Lightweight
Design
Explore how NaturalMSEQueries has
successfully applied to the several projects,
projects, showcasing its potential for lightweight
lightweight design in materials science.
2 User-FriendlyApproach
• Pioneer on the intersection between
between SWT and MSE
• Understand how our approach enables
enables domain experts to query
materials science data more effectively,
effectively, improving the overall
usability and accessibility of Semantic
Semantic Web technologies.
3 FutureDevelopment
NaturalMSEQueries + LLM (work in progress)
https://github.com/Mat-O-Lab/KnowledgeUI
andre.valdestilhas@bam.de