A framework for hypothesis generation and knowledge discovery in medical domain literature.

摘要

The vast amount of information in medical domain and health sciences collected in databases such as MEDLINE is growing rapidly. There has been increased interest in discovering the so-called new public knowledge from such databases. Swanson proposed an approach, called the ABC model, for mining Undiscovered Public Knowledge (UPK) in medical literature. Since its introduction, several attempts have been made in adopting and using the ABC model. Extensibility would be a key feature making it easier to develop future extensions. Noting the increased interest in using the model, we investigate properties of a desired framework which can be easily extended and enhanced. The exploratory nature of UPK discovery requires that the data mining tools be interactive and flexible. Also, the large amount of data to be processed needs to be handled efficiently. We identify three basic requirements: flexibility, extensibility, and interactivity, and show they can be realized by taking advantage of the pipes and filters architecture. The efficiency of our framework is due to allowing concurrent execution of multiple threads, provided as an additional benefit of our architectural design. We have designed and implemented a running prototype, ExaminMED, which provides various features such as possibility of choosing filters, adding or removing terms, and comparing and combining the results of various searches. The proposed framework has essential ingredients as an effective tool for UPK discovery.