An approach in adaptive inference engine for rule-based consultation systems
Duisburg (2006), IV, 115 Bl.
Dissertation / Fach: Elektrotechnik
Fakultät für Ingenieurwissenschaften » Elektrotechnik und Informationstechnik
Today, expert systems are widely used in business, science, engineering, agriculture, manufacturing, medicine, video games, and virtually every other field. In fact, it is difficult to think of a field in which expert systems are not used today (Lozano-Perez T., Kaelbling L., 2003). In the field of consultation, expert system has been applied very early, for example, MYCIN (Buchanan, B. G. and Shortliffe, E. H., 1984) can be seen one of the earliest applications of the expert system. Although there are many commercial products of expert system shells that can be applied to build consultation systems, they have shown some drawbacks: Accepting only one determined language for knowledge representation. Therefore, once one decided to use an expert system shell, it is not easy to make a change of it later because of having to edit the whole knowledge base in the new language again. Being so passive in comparison with human being. They can only do exactly what the knowledge engineer specified in the knowledge base. Therefore, it requires a lot of efforts for preparing knowledge base Using system resource is not optimal, when applying them to consultation systems. To deal with the above problems, this dissertation is aimed at presenting an approach of an adaptive inference engine for rule-based consultation systems, which is a traditional inference engine with some additional abilities: Being able to learn different languages for knowledge representations through training. Being able to find out by itself which question should be asked next without any interference from human being. The “Matching” can recognize which information is required for the reasoning, then it creates new rules for them and adds the rules into its knowledge base. Being able to find out by itself an optimal reasoning strategy (forward chaining, backward chaining or a mixture of both). It does not need any setting from human being for the reasoning strategy. The “Matching” determines it from the beginning and at run-time. Being able to find out syntax errors in rules. This is the evidence that the inference engine understands its rules and understands what it is doing as well. Being able to learn from experience to improve its performance Beside concepts and implementation of the adaptive inference engine, a mathematical evaluation on the effectiveness of the new matching algorithm is also presented.