Knowledge acquisition (KA) is a methodology of obtaining the knowledge of special domain from the domain expert, and knowledge based system (KBS) is an intelligent computer program that uses knowledge and inference procedures to solve problems that are difficult enough to require significant human expertise for their solutions, such as disease diagnosis, investment prediction, or computer science.
Meanwhile, KA is one of the critical bottlenecks in developing a KBS for obtaining the knowledge. Traditionally, knowledge engineers retrieve knowledge from human experts by interviewing, and then a computer can be applied to perform the similar problem as human experts do in the real world. In order to facilitate the efficiency, knowledge engineers draw support from technical documents to interact with human experts; therefore, the idea of ontology helps knowledge base system overcome several bottlenecks such as knowledge representation, sharing and reusing. However, in the real world environment, there is not only static knowledge but also dynamic knowledge which is intensely fickle. With the changing environment as time goes on, new objects in many domains are incrementally evolved or developed due to the explosion of knowledge. It results in the creation of new knowledge due to the new evolved objects. Avoiding a downward efficiency spiral involves constantly hatching latest information while trying to defend an existing knowledge line. Hence, knowledge can be classified as static knowledge and dynamic knowledge according to the stability of knowledge in dynamic environment over time. The static knowledge remains the same in the changing environment as time goes on. Since the different environment can change over time, the original knowledge constructed at a time may
the knowledge will be updated or evolved over time due to the adaptation of the changing environment. The knowledge evolution we proposed in this thesis is the iterative process to acquire evolutional knowledge in the changing environment.
Traditional KA methodologies, which are capable of acquiring static knowledge, can be classified into interviewing, machine learning and knowledge acquisition systems; however, the dynamic knowledge acquisition has hardly been discussed.
Knowledge engineers directly retrieve domain knowledge by interviewing with human experts, and transform the knowledge into the computerized format to help experts solve difficult problems in the real world. In general, to acquire dynamic knowledge, the experts are required to be aware of the occurrence of new objects in the interviewing approach and knowledge acquisition systems. However, it is still difficult for experts to be aware of the new object without any additional related information. The machine learning approaches, which can learn the useful knowledge of static objects according to the selected training cases, are usually lack of the ability of discovering new objects without new cases of dynamic objects in the training process. As we know, many KA systems and tools such as NeoETS [5], ACQUINS [6], KITTEN [24], EMCUD [17], KSSO [14], have been proposed to rapidly build prototypes and improve the quality of the elicited static knowledge of well-known objects by domain experts with/without knowledge engineers in the past twenty years.
However, most of them lack the ability of incrementally acquiring dynamic knowledge since the experts may not be aware of the occurrence of new objects without sufficient information.
EMCUD (Embedded Meaning Capturing and Uncertainty Deciding) [Hwang, 1991] was proposed to elicit the embedded meanings of knowledge (embedded rules bearing on m objects (O1, O2, …, Om) and k object attributes) following repertory
grids principles, which represents the information that domain experts take for granted but are implicit to the people who are not familiar with the application domain, and guide experts to decide the certainty degree of each embedded rule for extending the coverage of the original rules generated by acquisition table. Since the relative importance of each attribute to each object could be represented as attribute ordering table (AOT) in EMCUD, some minor attributes can be relaxed or ignored to capture the embedded meanings with acceptable CF. Assume some objects in O1 class, which are classified by original rules of O1, belong to the original object class (OO1) of O1; the other objects in O1 class, which are classified by embedded rules of O1, belong to the extended object class (EO1) of O1. However, some embedded rules may be with marginally acceptable certainty factor (CF) values due to the weak suggestions of domain experts. In the age of the knowledge explosion, some objects might be evolved with the times and could be classified by the embedded rules of O1 with weak CF values since some related ambiguous attributes (minor attributes) are ignored to classify these new evolved objects into O1 class. Although EMCUD extends the ability of KA system to classify a new object into an original object class with the weak embedded rules with lower CF values, it still lacks the ability to incrementally discover the new objects and integrate the corresponding dynamic knowledge of new objects into original knowledge base without rerunning the whole KA procedure to generate the knowledge. Moreover, the human experts are unlikely to be aware of the occurrence of the new evolved objects due to the lack of sufficient relevant information about the new objects.
In this thesis, we will propose a new collaborative knowledge acquisition method (CAKE – Collaborative Acquisition for Knowledge Evolution) to collect useful evidence through monitoring the frequent inference behaviors of weak embedded
rules and tracing trend events of objects with time in order to assist experts to efficiently adapt the certainty factor of dynamic knowledge of new objects according to the sufficient context information. In order to discover the new object, the VODKA (Variant Object Discovering Knowledge Acquisition) is proposed to facilitate the acquisition of new knowledge of variant objects by monitoring the frequently fired weak embedded rules. Since the evidence of object evolution may appear diversely in unpredictable time, a time interval tracing oriented mechanism, Trend Event Acquisition (TEA) [22], for constructing dynamic knowledge of new objects is proposed to adapt knowledge to current time by recording each interested attribute’s information in each time interval and update evolutional knowledge base. The VODKA generates a new acquisition table of new object, and the TEA generates a dynamic AOT table for capturing the evolutional embedded meaning of these objects.
However, some dynamic knowledge might be invisible in each KBS with VODKA and TEA. Several heuristics are proposed to assist experts in discovering the new evolved objects, including service-sensitive and symptom-similar heuristics. The context information, including static profile and dynamic behaviors, is designed to assist experts to be aware of the occurrence of dynamic objects according to service-sensitive heuristic and symptom-similar heuristic by collecting sufficient information of multiple sensors. The service-sensitive heuristic can help experts to discover similar behaviors result in different symptoms due to different individual or environment configurations. The symptom-similar heuristic can assist experts to recognize different behaviors result in similar symptoms between multiple sensors due to the polymorphic configurations. We think the legacy knowledge acquisition methods are inefficient to acquire the evolutional knowledge because it is unable to learn the evidences of variation and development. Finally, we will propose the
Dynamic EMCUD (D-EMCUD) to integrate new dynamic knowledge into original knowledge base.
Based upon the CAKE framework, a worm detection prototype system with the CAKE module is implemented to evaluate the effectiveness of integrating the new evolved knowledge into original knowledge base. Based upon the collaborative framework, the dynamic knowledge of new evolved objects could be elicited to discover the new variant worms generated by the attacking traffic generator in the experimental environment.