From: A hybrid method based on semi-supervised learning for relation extraction in Chinese EMRs
Algorithm 1: Bootstrapping | |
---|---|
Require: Labeled seed set L | |
Require: Unlabeled set U | |
Require: Reliable set N | |
Require: Threshold | |
   repeat: | |
      Train a single relation extraction model on L | |
      Run the relation extraction model on U | |
      Find (at most) N instances in U that the probability predicted by the relation extraction model is greater than \(\lambda\) | |
      Add them into L | |
   Until No data points available in U |