Table 4.

The 10-fold cross-validation Micro-$F_1$ score on all methods.

Fold	Methods-Other Teams		Methods-Our Works							Average/Fold
	Plain SVM [10]	Hierarchical SVMs [10]	BOW-BiGRU-Att	Word2Vec-BiGRU-Att	FastText-BiGRU-Att	GloVe-BiGRU-Att	ELMo-BiGRU-Att	FT-GPT-FC	FT-BERT-FC
F-1	0.682	0.739	0.658	0.710	0.728	0.719	0.750	0.743	0.7891	0.724
F-2	0.671	0.698	0.650	0.699	0.701	0.704	0.724	0.722	0.755	0.702
F-3	0.639	0.682	0.643	0.677	0.673	0.680	0.707	0.721	0.750	0.686
F-4	0.693	0.743	0.669	0.724	0.737	0.727	0.745	0.768	0.778	0.732
F-5	0.658	0.721	0.645	0.681	0.712	0.691	0.722	0.730	0.762	0.702
F-6	0.677	0.728	0.662	0.700	0.680	0.703	0.731	0.735	0.771	0.710
F-7	0.642	0.690	0.631	0.686	0.719	0.695	0.712	0.721	0.753	0.694
F-8	0.669	0.729	0.660	0.712	0.723	0.719	0.736	0.744	0.776	0.719
F-9	0.690	0.735	0.668	0.703	0.718	0.702	0.749	0.747	0.791	0.723
F-10	0.678	0.723	0.649	0.681	0.691	0.677	0.721	0.730	0.762	0.701
Average/Method	0.670	0.719	0.654	0.697	0.708	0.702	0.730	0.736	0.769	/

¹ The FT-BERT-FC gets the best performance with the bold number in each fold.