Skip to main content
. 2019 Jun 24;26(11):1272–1278. doi: 10.1093/jamia/ocz072

Figure 1.

Figure 1.

We train a neural network to predict billing codes given the text of clinical encounters. After the training is finished, we save the model. We use the saved model as a text encoder to create dense representations for the notes in a target task. These representations can be used to train a classifier.