Clustering is the technique to partition data patterns into different clusters. In real datasets, there are often a large number of unlabeled data patterns and a small number of labeled data patterns. Clustering technology can group data by using the unlabeled data only, but the accuracy is often poor. Classification can learn a classification model through labeled data. Classification can group data too. The classification accuracy is relatively high, but it needs a rich supply of labeled data. Therefore, in order to solve such partition problems of hybrid dataset, semi-supervised clustering is proposed. This paper proposes an improved artificial bee colony (ABC) semi-supervised clustering algorithm. The new algorithm tries to use both labeled data and unlabeled data. It is implemented by simulating the feeding of bees in nature. Experiment results show that the new clustering method performs very well, and it can improve cluster accuracy greatly.