“…Although manually labeled instances can make up for the lack of labeled instance to a certain extent, this process is time-consuming and laborious. Then, the semi-supervised learning (SSL; Zhou et al, 2003 , 2014 ; Zhu et al, 2003 ; Chapelle et al, 2006 ; Zhu, 2008 ; Zhu and Goldberg, 2009 ; Gao et al, 2010 ; Zhou and Li, 2010 ; Zhao and Zhou, 2018 ; Tao et al, 2019 ; Wang Q.-W. et al, 2019 ; Wang T.-Z. et al, 2019 ; Zhang et al, 2019c ) technique was proposed, which learns a model from a small amount of labeled instances and a large amount of unlabeled instances and solves the problem of insufficient labeled instance (i.e., poor generalization of the model obtained by supervised learning and inaccurate models obtained by unsupervised learning).…”