Biomedical Text Classification Using Augmented Word Representation Based on Distributional and Relational Contexts

Table 10

CNN classification performance using embeddings over the BioText Berkeley dataset.

Embeddings	Accuracy
	,		,		,		,
	Training	Validation	Training	Validation	Training	Validation	Training	Validation
GloVe_C	0.8887	0.8592	0.9177	0.8211	0.8907	0.8592	0.9073	0.8651
GloVe_W	0.8858	0.8563	0.9047	0.8123	0.8864	0.8334	0.9021	0.8270
GloVe_Merged	0.8487	0.8475	0.8584	0.8006	0.8493	0.8358	0.8643	0.8211
CE	0.8873	0.8646	0.9255	0.8323	0.8788	0.8328	0.9191	0.8894
WE	0.8749	0.8652	0.9201	0.8423	0.8688	0.8428	0.9285	0.8318
Merged	0.8507	0.8440	0.9155	0.8294	0.8516	0.8152	0.8862	0.8465

Bold means the best performance in the case of each dataset.