Semisupervised Learning Based Disease-Symptom and Symptom-Therapeutic Substance Relation Extraction from Biomedical Literature

Table 4

The results obtained with Tri-Training on the disease-symptom test set. Method 1 integrates three classifiers (the feature kernel, graph kernel, and tree kernel) with the same weight while Method 2 integrates them with a weight ratio of 4 : 4 : 2.

Method			-score	AUC
Feature kernel	83.00	80.08	81.51	88.80
Graph kernel	77.74	85.94	81.63	89.80
Tree kernel	57.38	94.14	71.30	76.00
Method 1	79.79	87.89	83.64	91.57
Method 2	79.93	85.55	82.64	90.75