Objective Based on the requirements of knowledge learning and application in the domain of biomedical, a kind of entity disambiguation algorithm is proposed to solve the problem of word ambiguity in entity recognition. Methods Entity disambiguation is usually divided into two parts: candidate generation and entity disambiguation. In this paper, candidates of name mention are generated based on the knowledge base method, and candidate entities are filtered based on the prior probability in the knowledge base of the candidate entity, which ensures the recall rate of the candidate entity set and the noise reduction in the disambiguation stage effectively. In the stage of entity disambiguation, we propose a disambiguation method based on the contextual characteristics of the entity, construct probabilistic model to compute the similarity between entity context and entity reference context, and select the largest similarity entity as the target entity. Then, we conduct entity disambiguation experiments for name mentions which are recognized from the literature, and determine the correctness of entity disambiguation by domain experts. Finally, we compare the accuracy of entity disambiguation under different algorithms. Results The accuracy of the proposed method is 83% with our dataset, which is higher than that of other algorithms. Conclusions The entity disambiguation algorithm based on context features is the best in the field of entity disambiguation.