Given　that　citations　are　not　equally　important,　various　techniques　have　been　presented　to　identify　important　citations　on　the　basis　of　supervised　machine　learning　models.　However,　only　a　small　volume　of　data　has　been　annotated　manually　with　the　labels.　To　make　full　use　of　unlabeled　data　and　promote　the　learning　performance,　the　semi-supervised　self-training　technique　is　utilized　to　identify　important　citations　in　this　work.　After　six　groups　of　features　are　engineered,　the　semi-supervised　versions　of　SVM　and　RF　models　improve　significantly　the　performance　of　the　conventional　supervised　versions　when　un-annotated　samples　under　75%　and　95%　confidence　level　are　rejoined　to　the　training　set,　respectively.　The　AUC-PR　and　AUC-ROC　of　SVM　model　are　0.8102　and　0.9622,　and　those　of　RF　model　reach　0.9248　and　0.9841,　which　outperform　their　counterparts.　This　demonstrates　the　effectiveness　of　our　semi-supervised　self-training　strategy　for　important　citation　identification.　©　2021　CEUR-WS.　All　rights　reserved.