ChEMU consists of information extraction tasks over chemical reactions described in patents (https://chemu-patent-ie.github.io/). The goal of the task 1 (NER) was to identify chemical compounds and other entities involved in chemical reactions (like time, temperature, yield) in patents text and to assign a type to each entity according to its role in the context of the reaction.
The model developed by the DeST team (lasige_BioTM) outperformed the baseline model and was the third best result:
Exact evaluation
| Precision | Recall | F1-Score | |
| Baseline | 0.9071 | 0.9071 | 0.8893 |
| DeST | 0.9327 | 0.9457 | 0.9392 |
Relaxed evaluation
| Precision | Recall | F1-Score | |
| Baseline | 0.9208 | 0.9504 | 0.9354 |
| DeST | 0.9583 | 0.9960 | 0.9768 |
