Table 3.

Feature set description

Feature	Description
Bag of Words	Bag of Words in a 5-word window.
Part of Speeches	Part of Speeches in a 7-word window.
Capitalization	Convert all alphabetic characters of the words to uppercase [31]. The window size is 5.
Case pattern	The patterns are generated by the following steps. Similar to [32], any uppercase alphabetic character is replaced by “A” and any lowercase one is replaced by “a”. In the same way, any number is replaced by “0”. The window size is 3.
Word representation	We use word2vec to acquire 700 clusters from the unlabeled clinical narratives and give each cluster a different serial number. Then we take the serial number of the clusters as a feature. The window size is 3.