In order to detect the concept automatically in the specific field to obtain knowledge unit and relationship to create a knowledge map. This paper proposes a method based on conditional random field and information entropy, which divides the problem of concept discovery into two parts: the concept recognition and the new concept discovery. Firstly, the boundary of the text sequence is forecasted by the conditional random field. Compared with the concept in the dictionary, the candidate of the new concept is selected, the approximate position of the concept is found and the conceptual internal consistency is judged by mutual information. Determining the concept of boundary freedom is to carry out concept discovery by information entropy. Finally we can get the new professional concepts of the field of construction project. Experiments show that this method can effectively improve the accuracy, recall rate and efficiency of the concept identification and discovery in the field of specialization. It is also an effective alternative to the cascaded conditional random fields.