Figure 6.

Training procedures for TextRNN-based models with different batch sizes whose $\mathrm{F}{1}_{micro}$ are in bold in Table 3, Table 4 and Table 5.