This repository contains code to train and evaluate German medical embeddings. Example code and usage examples are provided in Jupyter notebooks (1 Vectorization, 2 Evaluation).

Define relevant paths in config.json (have a look at the default file).

To extend the processing pipeline do the following:

Extend vectorization/embeddings.py->Embeddungs.sentence_data2vec() before calling Embeddings.calculate_vectors().

Extend vectorization/embeddings.py->Embeddings.calculate_vectors() by new embeddings algorithm which accepts sentences as input.

Add new ressources to resource/other_resources.py or resource/UMLS.py by inheriting the abstract class Evaluator and implementing its abstract methods.

Add new benchmarks to benchmarking/benchmarks by inheriting the abstract class Benchmark and implementing its abstract methods. Use the constructor to define relevant resource files such as knowledge bases, which will be passed by evaluate_embeddings.py or other running instances as list.