The invention discloses a method, a device and a system for information classification based on Bayesian structure learning. The method comprises the steps: extracting information according to a material provided by a user, arranging the information into a structured data list, obtaining an incidence relation of all diseases through Bayesian structure learning according to a historical user data set and an information database, constructing a Bayesian network structure, aggregating the front and back sequence diseases belonging to the same type in the Bayesian network to obtain a corresponding disease classification, and forming a corresponding strategy scheme according to an expert experience knowledge graph to output an analysis result to the user. The method exacts information from materials such as user medical records based on Bayesian structure learning, performs structured processing on the extracted data, obtains association relationships of each disease in combination with a historical data set and an information database and forms a corresponding reference strategy scheme in combination with the expert experience knowledge graph, and the scheme can be important reference information to assist doctors in judgment, so as to save processing time.