Table 1.

Predictive performance by algorithm

	Algorithm	LSTM	ELMo	BERT
Full dataset (434 labels)	Top-1	0.63	0.63	0.65
	Top-2	0.77	0.78	0.80
	Top-3	0.84	0.85	0.87
	Top-4	0.88	0.88	0.90
	Top-5	0.90	0.90	0.92
Reduced dataset (188 labels)	Top-1	0.66	0.66	0.69
	Top-2	0.81	0.81	0.84
	Top-3	0.88	0.88	0.90
	Top-4	0.90	0.91	0.93
	Top-5	0.93	0.93	0.94