Abstract: The popularization of electronic clinical medical records (EMRs) makes it possible to use automated ways to quickly extract high-value information from EMRs.As a kind of crucial medical information,tumor medical event is typically composed of a series of attributes describing malignant tumors.Recently,tumor medical event extraction has become a research hotspot in the academic community,and many influential academic conferences publish it as an evaluation task and provide a series of high-quality manually annotated data.Aiming at the discrete characteristic of tumor event attributes,this paper proposes a joint extraction method,which realizes the joint extraction of tumor primary site and primary tumor size and also the extraction of tumor metastasis sites.In addition,aiming to alleviate the small counts and types of annotated tumor medical texts,this paper proposes a pseudo-data generation algorithm based on the global random replacement of key information,which improves the transfer learning ability of the joint extraction method for different types of tumor events.The proposed method wins the third place in the clinical medical event extraction evaluation task of CCKS2020,and extensive experiments on CCKS2019 and CCKS2020 datasets verify the effectiveness of the proposed method.

Key words: Chinese electronic medical record, Joint extraction, Medical event extraction, Transfer learning, Tumor event