| Summary: | 博士 === 國立臺灣科技大學 === 營建工程系 === 95 === In the process of knowledge accumulation, due to the difference of knowledge source in the knowledge base and opinions of experts, the knowledge in the knowledge base would have inconsistency with the same meaning, conflict and data, which may lead to an improperness of knowledge use as a result of change of time, new technology, new regulation, new methodology and new evidence and other factors. Rule-based knowledge base expert systems have traditionally emphasized the verification of structural errors in the rule base. For conflicting or overlapping rules, designated rules are usually followed to implement prioritized or direct deletions. However, there exist no proper methods by which to resolve conflicts, inconsistencies or redundancies in value. Due to the uncertainty of uncertain knowledge itself, it is difficult to treat conflicting rules, and the citation of erroneous knowledge leads to mistakes in decision making. Among users, 94% report perplexity when conflicting or redundant rules are cited. It is therefore a necessity to confirm the existence and reliability of the cited knowledge.
This study proposes integrates methods of conditional probability, and vector matrices to establish a conditional probability knowledge similarity algorithm and calculation system. This calculation system can quickly and accurately calculate rule-based knowledge similarity matrices and capture the conflicting rules and redundant rules. For the conflicting rules and redundant rules, the current study attempts to provide an uncertainty rule-based knowledge conflict treatment inference model by integrating a group decision and an uncertainty inference. In the model, a “reliable factor” refers to the reliability level of the knowledge containing conflicts, redundancies or inconsistencies in values, while the “certainty factor” indicates the existence of the knowledge itself. A “certainty reliable index” is used to show both the existence of the knowledge itself and its reliability. Based on knowledge relationship, a rule-based uncertainty knowledge value-adding treatment inference model is established to perform value-adding treatment such as merge, integration, innovate, search, delete, and add, so that certainty reliable indexes of the rules can be obviously represented, For conflicting or overlapping knowledge, it is suggested that the knowledge with a higher certainty reliable index be chosen. Among users, 92% reported that the model is helpful to knowledge application and an aid to the decision-making process. It can more effectively prevent mistakes in decision making and enables users to acquire more benefits from the knowledge application.
|
|---|
