While current information technologies allow to collect large data sets, using the archived data is often complex, time consuming, or simply impossible. For example, less than millions of reports on medical device malfunction, patient injury and death are collected every year in the Manufacturer and User-Facility Device Experience (MAUDE) database. An adverse event is manually classified according to a hierarchical coding system. The current coding system uses an arbitrary taxonomy, which causes redundant, too general, or too specific categories. Fixed categories fail to house the unprecedented observations, resulting in loss of information. Departmentalized categories tend to prevent holistic understanding of a problem. In an effort to meaningfully organize archived data, this study suggests an analytic scheme to organize categories hierarchically. Principles and relevant measures based on the theories of category representation are proposed. Also, text mining algorithm implements flexible restructuring of categories with minimal human intervention, facilitating automatic updates to maintain adaptability to unprecedented observations. To demonstrate the benefit of the proposed categorization scheme, a case study is conducted is to generate taxonomy based on a set of retraction problems in MAUDE. Results have implications for the benefits of using flexible taxonomy to categorize archived data.