Table 5.

Performance of classification models using only lexical features according to different evaluation metrics for the task of annotating caregiver interview session transcripts. Highest value for each metric and codebook size across all models is highlighted in boldface.

Cls.	Model	Acc.	Prec.	Rec.	F1	Kappa
16	NB	0.571	0.608	0.571	0.575	0.518
	NB-M	0.633	0.629	0.633	0.604	0.573
	J48	0.578	0.563	0.578	0.567	0.514
	AdaBoost	0.602	0.582	0.602	0.588	0.539
	RF	0.640	0.631	0.640	0.596	0.574
	DiscLDA	0.482	0.442	0.482	0.421	0.362
	CNN	0.657	0.641	0.657	0.648	0.512
	SVM	0.664	0.653	0.664	0.639	0.606
19	NB	0.477	0.504	0.477	0.467	0.434
	NB-M	0.536	0.539	0.536	0.512	0.487
	J48	0.436	0.431	0.436	0.432	0.382
	AdaBoost	0.467	0.457	0.467	0.460	0.415
	RF	0.507	0.508	0.507	0.467	0.450
	DiscLDA	0.374	0.370	0.374	0.333	0.287
	CNN	0.510	0.498	0.510	0.504	0.401
	SVM	0.545	0.547	0.545	0.535	0.497
58	NB	0.379	0.392	0.379	0.370	0.350
	NB-M	0.442	0.404	0.442	0.386	0.401
	J48	0.340	0.321	0.340	0.328	0.302
	AdaBoost	0.381	0.359	0.381	0.366	0.344
	RF	0.402	0.358	0.402	0.352	0.358
	DiscLDA	0.288	0.258	0.288	0.234	0.229
	CNN	0.118	0.102	0.118	0.109	0.032
	SVM	0.451	0.420	0.451	0.418	0.414