This paper discusses techniques that improve the performance of a case retrieval system, after it is deployed, as a result of the continued usage of the system, by remembering previous episodes of question answering. The user generates a request for information and the system responds with the retrieval of relevant case(s). A history of such transactional behavior over a given set of data is maintained by the system and used as a foundation for adapting its future retrieval behavior. With each transaction, the system acquires information about the usage of the system that is subsequently used to adjust the behavior of the system. This notion of a case retrieval system draws on a distinction between the system in isolation and the system as it is used for a particular set of cases. It also draws on distinctions between the designed system, the deployed system, and the system that emerges as it is used.
Registration: ISBN 978-0-262-51091-2