A visual and interactive environment for knowledge acquisition
Hotz, Kenneth Joseph
Knowledge acquisition continues to be a major bottle-neck in the development of knowledge based systems. The process of identifying, representing, and verifying knowledge has been attacked from many different perspectives. Various tools have been created to automate portions of the process of interviewing or interacting with an expert. Many of these methods focus on guiding the expert in identifying knowledge which serves a specific purpose, often at the expense of forcing the expert into an awkward thought process. Other approaches view acquisition as a separate and distinct task where the tasks of representing and verifying the knowledge are performed separately. The time to perform the entire process can be significant causing problems of recollection should errors or inconsistencies exist in previously presented knowledge. To overcome these problems knowledge acquisition should be seen as a complete and integrated series of steps which present the expert with the ability to directly manipulate the knowledge in a manner which seems appropriate. This thesis presents a model for knowledge acquisition, called VENI (I KAME - Interactive Knowledge Acquisition Methodology and Environment), which defines an interactive environment consisting of a set of tasks sufficient for identifying, representing, and verifying knowledge. The basis for this model is not to direct the expert through systematic questioning. The model allows an expert to interact directly with the knowledge and react, in 'real time', to knowledge supplied by integrating any new, deleted, or modified knowledge with the existing knowledge base. Knowledge is acquired or modified only when explicitly directed by the expert. The acquisition process is based on an iterative cycle of: examining knowledge relevant to a presented case; getting explanations of the presented knowledge, making changes/extensions to the knowledge, and analysing the impact of those changes/extensions. In eliminating an interviewing component, the expert is allowed to put whatever knowledge is deemed appropriate without being forced into answering specific questions. To assist the expert, VENI provides two types of support. One, all communication of the existing knowledge, and manipulation of the knowledge base is done through the use of graphical images and direct manipulation of those images. Second, a set of facilities is defined which provide explanations of the relevant knowledge to assist the expert in making decisions about how to extend or change the knowledge base so that the presented case is adequately explained. A limited implementation of key components is described where choices for specific components are made which show the ability of the components to achieve the purposes defined for them in the general model. The interaction and behaviour of the components is further demonstrated through the use of case examples. To demonstrate the applicability of the model, a small-scale evaluation was conducted. In this evaluation an independent group of potential users were given a set of cases to explain within VENI's environment. After only a short training period and short working session, each evaluator was able to construct a small knowledge base, capable of solving a small set of problems in the domain of exercised induced bronchospasm and asthma. While not an extensive evaluation, the results support the claim that VENI can assist in developing a knowledge base.