Generative Knowledge Representation is one of the common knowledge representation methods. It is based on a large number of causal relationships between various knowledge in the human brain memory model and expressed in the form of "if-then", that is, generative rules. This form of Rule captures the behavioral characteristics of humans to solve problems, and solves problems through understanding the cyclical process of action.
A production system consists of a rule repository, a comprehensive database, and a control organization.
Generative rule notation has obvious advantages:
- Good naturalness. The generative expression represents knowledge in the form of "if-then", which is basically consistent with the judgment knowledge of humans, intuitive, natural, and easy to reason;
- Except for the general structure of the system, the interaction methods of each part, and the representation of rules, there are no specific provisions on other implementation details of the system, this is because designers have great flexibility in developing practical systems and can adopt appropriate implementation technologies as needed. In particular, they can introduce various heuristic knowledge that is meaningful to solving problems into the system;
- The format of the representation is fixed, the format is single, and the rules are independent of each other. The whole process is only the matching of the front part and the action of the back part. The information provided by the matching is only successful and failed. There is generally no recursion in the matching, and there is no complicated calculation, so the system is easy to establish;
- Because the knowledge in the rule repository has the same format and the global database can be accessed by all rules, the rules can be processed in a unified manner;
- Good controllability. generative rules are the most basic unit of knowledge in a rule. rules can only be associated with each other based on the volume of global data, it facilitates the addition, deletion and modification of knowledge;
- Generative notation can represent both deterministic knowledge units and deterministic knowledge. It can represent both heuristic knowledge and procedural knowledge. It can represent domain knowledge, it can also represent metadata.
However, generative rule notation also has the following Disadvantages:
- Low reasoning efficiency: Because the knowledge in the rule repository is in a unified format, and the association between rules must be based on a global database, the reasoning process is a process of repeated "matching-elimination of conflicts-execution. In addition, in each Reasoning Cycle, we must constantly search for the conditions of all rules and perform pattern matching. In principle, this approach will inevitably reduce the reasoning efficiency, as the scale increases, the disadvantages of low efficiency will become more and more prominent, and even the combination explosion problem may occur.
- Not intuitive: databases store independent rules, and the relationships between them are hard to be viewed in an intuitive way;
- Lack of flexibility: the knowledge of generative expressions has a certain format, and rules cannot be called directly. Therefore, it is difficult to express the knowledge with structural or hierarchical relationships. And cannot provide flexible explanations.
The generative method is the preferred Knowledge Representation Method for expert systems. The dendral system used by the chemical industry to determine the molecular structure, the mycin system used to diagnose meningitis and blood virus infection, and the prospector system used to estimate mineral deposits are examples of Knowledge Representation and reasoning using this method.
