Forschungsthemen
[MA] Design of a NLP Framework for Terms Mapping to Tree-Based Taxonomies
The objective of this thesis is to develop a generalized NLP software framework that can accurately map given terms to a given tree-based taxonomy. The proposed approach will use stacked transformers for hierarchical classification based on Open Source Machine Learning technology. The proof of concept and its performance measurements will be based on a medical use case including example data (MedDRA). The desired framework should be designed w.r.t. to certain software qualities, such as transferability, extendibility, reusability, explainability and robustness, hence it should not be limited to any particular domain or taxonomy/ontology, and can be adapted to work with different types of knowledge models and technologies.
The task incorporates with other student works and projects of the AI4Hematology working group and will be part of a larger software platform to tackle data-driven tasks and issues in the domain of Digital Health.
Betreuer: Karsten Wendt