Jointly detecting and classifying outcomes within biomedical text records. Experiments and results of training this model can be found in this paper.

LCAM simulataneously extracts outcomes (Outcome span detection (OSD)) and classifies them (Outcome Classification (OC)) given an input sentence, e.g.

There was no significance between group differences in the incidence of wheezing or shortness of breath.

OSD detects and extracts the outcomes ''Wheezing'' and ''shortness of breath'', and OC classifies them as ''Physiological outcomes''.

Download processed versions of datasets EBM-COMET and EBM-NLP from here

Label document alignment (L-D-A)

Part of this work is introducing a re-usable unsupervised text-alignment approach that extracts parallel annotations from comparable datasets. LDA implementation here. We leverage L-D-A to merge the two datasets above and evaluate LCAM on EBM-COMET+EBM-NLP. Preprocessed version of this dataset is downloadable here.

Using BiLSTM as a Text encoder

python train.py --data multi-labelled-data/ebm_comet/lcam --vocabularly multi-labelled-data/lcam_vocabularly/ --abs_encoder --attention
(--attention for the attention layers and --abs_encoder for incoproating abstract representation)

Using BioBERT as a Text encoder

python train_transformer.py --data multi-labelled-data/ebm_comet/lcam --vocabularly multi-labelled-data/lcam_vocabularly/ --transformer ../pretrained_transformer_models/biobert/ --layer -1

@article{abaho2021detect,
  title={Detect and Classify--Joint Span Detection and Classification for Health Outcomes},
  author={Abaho, Michael and Bollegala, Danushka and Williamson, Paula and Dodd, Susanna},
  journal={arXiv preprint arXiv:2104.07789},
  year={2021}
}