Production Introduction

The term " production " was first proposed by the American mathematician Post in 1943. According to the string replacement rules , he proposed a calculation model called Post machine, and each rule in the model is called a production. Therefore, production representation is also called production rule representation .
In 1965, Newell and Simon in the United States established a cognitive model using this principle. In the same year, when Stanford University designed the first expert system DENDRAL, it adopted the structure of the production system.

production expression

A triple means: (object, attribute, value) or: (relationship, object1, object2)
Example: Lao Li's age is 40 years old: (Li, age, 40)
- Lao Li and Lao Wang are friends: (friend, Li, Wang)

Quads representing: (object, attribute, value, confidence) or: (relationship, object1, object2, confidence)
Example: Lao Li's age is likely to be 40 years old: (Li, age, 40, 0.8 )
- Lao Li and Lao Wang are unlikely to be friends: (friend, Li, Wang, 0.1 )

The difference between production and implication in predicate logic

Implications can only represent precise knowledge , and the logical value is true or false, while productions can also represent uncertain knowledge ; during reasoning, exact matching is not required .
The implication itself is a predicate formula and has a truth value , while the production has no truth value .

insert image description here

The rule base is used to describe the production set of knowledge in a certain field , including the transformation rules that transform the problem from the initial state to the target state (or solution state). It is the core of the expert system .

The comprehensive database is also called the fact library , which is used to store the initial facts input, the facts input from the external database , the intermediate results (facts) and the work area of the final results . is dynamic .

The reasoning engine is a program or a group of programs used to control and coordinate the operation of the rule base and the comprehensive database, including reasoning methods and control strategies . Decide what rules to choose or how to apply them.

Usually there are 3 steps from selecting a rule to executing an action: matching , conflict resolution , and action .

Matching : Matching is to compare the facts in the current comprehensive database with the conditions in the rules , and if they match, this rule is called a matching rule .
Conflict resolution : When there are multiple matching rules , the rule selected by the conflict resolution policy to be executed in the operation part is called the enabling rule . There are many conflict resolution strategies, among which specificity sorting, rule sorting, scale sorting and nearest sorting are common conflict resolution strategies.
Action : An action is the operational part of a rule that executes . After the operation, the current comprehensive database will be modified, and other rules may become enabled rules.

Forward reasoning : Starting from the known facts , the conclusion is obtained through the rule base .
Reverse reasoning : Starting from the goal (as a hypothesis) , use the rules in reverse to obtain known facts .
Two-way reasoning : Two-way reasoning is a way of reasoning that combines forward reasoning and reverse reasoning. The reasoning is carried out simultaneously from two directions until the results of the two directions on a certain intermediate interface match, and then it ends successfully.