This paper explores and defines a judicial knowledge reasoning model based on representation learning. The judicial judgment documents of our country collect data in xml format, and represent the key information of the case through the form of attribute-value. In the judicial documents, there is a specific relationship between the information. In order to describe this relationship and predict the element values in the xml through this relationship, this paper makes full use of the data characteristics and the theory of representation learning, and uses the judgment document data as the judicial knowledge base. Construct a judicial knowledge reasoning model based on representation learning, and help judicial staff to check for missing gaps and provide reference analysis when writing judgment documents.