. Author manuscript; available in PMC: 2018 Nov 1.

Published in final edited form as: J Biomed Inform. 2017 Jun 15;75 Suppl:S129–S137. doi: 10.1016/j.jbi.2017.06.014

Table 4.

Experimental performance of P@K for candidate psychiatric symptoms ranking by using sentence and phrase embeddings based sematic similarity. The employed corpora - psychiatric notes, psychiatric forum postings, and MIMIC II - for embedding generation are integrated incrementally. (%)

Method		P@5	P@10	P@100	P@200	P@1000
Baseline	Dictionary	100.0	100.0	45.0	26.0	13.0
Baseline	TF-IDF	80.0	80.0	52.0	35.5	17.6
Sentence embedding	notes	100.0	80.0	49.0	32.55	17.4
	+forum	100.0	80.0	61.0	45.5	26.3
	+mimic	100.0	80.0	53.0	37.5	18.0
Phrase embedding	notes	100.0	80.0	44.0	27.0	12.9
	+forum	100.0	80.0	51.0	33.5	20.0
	+mimic	100.0	80.0	52.0	31.5	16.1