Table 4.

Performance of classification models using only lexical features according to different evaluation metrics for the task of annotating adolescent 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
17	NB	0.544	0.603	0.544	0.552	0.497
	NB-M	0.670	0.662	0. 670	0.643	0.622
	J48	0.595	0.573	0.595	0.580	0.539
	AdaBoost	0.627	0.600	0.627	0.609	0.574
	RF	0.670	0.662	0.670	0.625	0.616
	DiscLDA	0.477	0.454	0.477	0.431	0.388
	CNN	0.678	0.633	0.678	0.670	0.509
	SVM	0.708	0.705	0.708	0.680	0.663
20	NB	0.487	0.509	0.487	0.482	0.448
	NB-M	0.579	0.582	0.579	0.559	0.537
	J48	0.479	0.467	0.479	0.470	0.431
	AdaBoost	0.504	0.488	0.504	0.493	0.458
	RF	0.563	0.564	0.563	0.519	0.514
	DiscLDA	0.400	0.410	0.400	0.356	0.330
	CNN	0.586	0.588	0.586	0.587	0.476
	SVM	0.610	0.611	0.610	0.592	0.571
41	NB	0.406	0.434	0.406	0.405	0.375
	NB-M	0.513	0.479	0.513	0.484	0.478
	J48	0.396	0.375	0.396	0.382	0.356
	AdaBoost	0.436	0.412	0.436	0.421	0.398
	RF	0.495	0.487	0.495	0.453	0.455
	DiscLDA	0.362	0.387	0.362	0.301	0.304
	CNN	0.396	0.369	0.396	0.382	0.170
	SVM	0.537	0.513	0.537	0.504	0.502