A KIF Based Knowledge Exchange Method

This paper is supported by National High Technology Research andDevelopment Program of China(Integrated Environment of IntelligentApplications, IEIA), our research area is knowledge sharing and knowledgeexchange. We study on the algorithms of the exchange method betweendifferent knowledge representations and different uncertainty models. Ourtask is to implement the knowledge base visiting service of the knowledgebase middleware of IEIA.In recent years, with the development of computer applications andnetwork technology, more and more knowledge resources appear in programs.Because the representations are not universal, we are not able to use theresources on the web and terminals effectively. How to organize and managethe knowledge and do the necessary exchanges to make the knowledge andinformation sharable and reusable becomes a hot research topic.We always use ontology when we do knowledge exchange. Ontology is aconcept model which can describe the system on a semantic and knowledgelevel. It can acquire knowledge from a specific field in a universal way, andprovide the common understandings about the concepts in the field. Weshould design standard ontology. We need some objective standards when wethink about how to design ontology to construct the concepts for knowledgeexchange and reusing. We introduce five ontology design standards in ourpaper: Clarity, Coherence, Extensibility, Minimal Encoding Bias andMinimal Ontological Commitment. The standards can guide and evaluate ourdesign and make sure that the ontology model can play a better role. Anexample is given to show the meanings and the relations of the fivestandards.Without the support of machine languages, computer can not transactknowledge. First Order Logic(FOL) is an important way to representknowledge. Knowledge Interchange Format(KIF) is a language based onFOL. KIF can be used for the translation between different computerprograms, it is built by Stanford University and it is a proposed draftAmerican National Standard. KIF has been widely used in expert system,database and intelligent agents as a bridge between two knowledgerepresentation languages. In order to make the knowledge sharing andknowledge exchange available and ease the development, many standardsand protocols are brought out, the importance of KIF is more and moreobvious.We have studied the translation method between KIF and Rule andProcedure Based Language(RPBL) in this paper. IEIA uses RPBL torepresent causal knowledge. Although RPBL has advantages to describemodels in production field, it also has some disadvantages with thedevelopment of knowledge engineering technology. IF we want to expand theuses of knowledge base, we need to communicate with other knowledgerepresentations. For this purpose, we analyze the syntax and semantics ofRPBL, use the logic understandable, exact FOL description method, constructKIF ontology, and finally define the translation relation from RPBL to KIF.XML serialization can make it easier to transact KIF ontology, and make theknowledge represented by KIF easy to use, transit and save, so we define abasic KIF/XML representation method.This paper has also discussed the transformation between the CertaintyFactor Model and Bayesian Model. These two models are the most useduncertain reasoning models in IEIA. During the applications of expert system,we usually do not have enough evidence to make decision, so there are manyuncertain facts. How to handle these uncertainties needs uncertain reasoningmodel. In the last several decades, many researchers have tried manymethods to represent and transact uncertainty, such as evidence theory,certainty factor model, PROSPECTER model and fuzzy set theory. In recentyears, more and more people focus on probability method. There are twoviews on the interpretation of probability, subjective probability and objectiveprobability. Subjective probability holders think that probability is a degree ofbelief, but objective probability holders are against subjective probability,they believe probability has nothing to do with belief. If different expertsystems use different uncertain reasoning models in a distributed expertsystem, it is necessary to transform the uncertainty of a proposition from onemodel to another when they cooperate to solve problems. In order toimplement the transformation of uncertainties between the certainty factormodel and the Bayesian model, we find that the key problem is how to obtainthe values of prior probabilities. This paper provides four methods toimplement the transformation. They are expert value method, evidenceinstance method, synthesize instance method and objective probabilitymodeling method. We accomplish the task from both the subjectiveprobability view and the objective probability view, so our work is moresystematic and trustworthy.Finally, as the application of the theory study result, the paper designsand implements the knowledge base transformation system between Bayesianknowledge base and rule knowledge base. These two knowledge base are themost frequently used in the expert systems of IEIA. The transformationarchitecture is composed of Bayesian knowledge base edit interface,knowledge exchange process and uncertainty factor model based ruleknowledge base edit interface. Knowledge engineers can make use ofhuman-machine interface to edit source knowledge base and targetknowledge base;the transformation involves rule extraction, KIFrepresentation, translation from KIF to RPBL and the transformation ofuncertainty. During the mapping process we need to finish two main tasks:the knowledge representation disposal and uncertainty disposal. We take KIFontology as our exchange standard, follow the methods we told before, wechange the inner knowledge structure from one form to another. In order toshow the macroscopic view of knowledge sharing, we provide a knowledgesharing architecture which describes the interaction of five roles: ontologyserver, network clients, remote applications, knowledge based programs andagents.