Algorithm 1: MvAl-MFP

Input: The initial labeled set $S_L$, the initial unlabeled set $S_U$, the number of labels L, the human expert $H$, the multi-label learning model $M$, the number of feature representation methods $V$, feature representation methods $\left\{B_1,B_2\dots B_V\right\}$, the number of selected unlabeled samples in each iteration $\lambda$. Output: Ensemble multi-label classifiers {$M_1,M_2\dots M_V$} Procedure:  For i = 1:V   Transform $S_L$ and $S_U$ by $B_i$;   Train the initial classifier $M_i$ on the transformed $S_L$;  End for  Repeat until satisfying the presetting stopping criterion     For i = 1:|$S_U$|      Generate ${\mathbf{Y}}_i^{\prime }$ by adopting {$M_1,M_2\dots M_V$} to predict its L labels (see Equation (3));      Calculate its voting entropy ${Entropy}_i$ by Equation (4);   End for   Rank voting entropies of all unlabeled samples in descending order, and select Top-λ ones into $S_U^{*}=\left\{s_1,s_2,\dots ,s_{\lambda }\right\}$ by Equation (5);   Submit $S_U^{*}$ to H for acquiring real labels by wet-lab experiments;   Add $S_U^{*}=\left\{s_1,s_2,\dots ,s_{\lambda }\right\}$ with real labels into $S_L$;   Remove $S_U^{*}=\left\{s_1,s_2,\dots ,s_{\lambda }\right\}$ from $S_U$;   Update {$M_1,M_2\dots M_V$} by using the extended $S_L$; Output the final {$M_1,M_2\dots M_V$} and make decision for future unseen samples by them (see Equation (6)).