Based on body learning, the invention relates to a Web service semantic extraction method, which comprises steps as follows: firstly, extracting semantic information of a current WSDL file: successively analyzing grammatical structure and type definition structure of each type by reading the type definition information of the WSDL file, and building a corresponding body structure; secondly, after completing the analysis of the current WSDL file, loading the obtained body into a database and completing a refining process to the body: successively extracting the body obtained after the analysis, searching the body library for the same or related bodies, acquiring implicit relationship between the bodies by direct comparison of the bodies, enriching the body structure, and after the completion of the process, loading the body and the comparison result into the database; and finally, building a relationship between service and the body by analysis of service-related information of the current WSDL file. The method can automatically add corresponding semantic information for service with non-semantic information by adopting the body learning method, and can be used in the service discovery field, so as to improve the efficiency of service discovery.