Expert evidence document recognition is the key step for expert search. Combining specialist candidate document independent page features and correlation among pages, this paper proposes an expert evidence document recognition method based on undirected graph model. First, independent page features such as words, URL links and expert metadata in all kinds of expert evidence document, and correlations such as links and content among candidate expert evidence document are analyzed. Then, independent page features and correlation among pages are integrated into the undirected graph to construct an undirected graph model for expert evidence document recognition. Finally, feature weights are learned in the model by using the gradient descent method and expert evidence document recognition is achieved by utilizing Gibbs Sampling method. The effectiveness of the proposed method is verified by comparison experiment. The experimental results show that the proposed method has a better effect.