Knowledge fragment enrichment aims to complete user input concept fragment by augmenting each concept with rich domain information. This is a widely studied problem in cognitive science, but has not been intensively investigated in computer science. In this paper, we formally define the problem of knowledge fragment enrichment in domain knowledge base and develop a probabilistic graphical model to tackle the problem. The proposed model is able to model the dependencies among concepts in the input knowledge fragment and also capture the probabilistic relationship between concepts and domain entities. We empirically evaluate the proposed model on two different genres of datasets: PubMed and NSFC. On both datasets, the proposed model significantly improves the accuracy of label prediction task by up to 3–9% (in terms of MAP) compared with several alternative enrichment methods.