Table 2.

Human-BERT group assignment agreement.

Ratings	Fleiss' kappa	SE	95% CI
			Lower	Upper
Overall	0.617	0.016	0.585	0.648
Group 1	0.636	0.028	0.582	0.690
Group 2	0.476	0.028	0.422	0.530
Group 3	0.589	0.028	0.535	0.643
Group 4	0.716	0.028	0.662	0.770

Overall and per-group Fleiss’ kappa values on agreement between six human raters and the BERT-based 4-group cluster assignments for 103 FTD items. Confidence intervals (CI) are asymptotic and indicate the precision of each kappa estimate.