Artificial Intelligence: Chapter II Knowledge Representation method _ Artificial Intelligence

Chapter II Knowledge Representation method

Teaching Content: This chapter discusses the various methods of knowledge representation, which is one of the three main contents of artificial intelligence course (knowledge representation, knowledge inference, knowledge application) and also the basis of studying other content of artificial intelligence.

Teaching emphases: State space method, problem attribution constitution, predicate logic method, semantic network method.

Teaching difficulties: State description and state space diagram, problem reduction mechanism, permutation and unity.

Teaching methods: Classroom teaching mainly, combined with the "discrete mathematics" and other already learned content real-time questioning, collecting students learning, make full use of the network curriculum multimedia materials to express the abstract concept.

Teaching Requirements: Focus on the use of state space, problem-constitution, predicate calculus, semantic network method to describe the problem, solve the problem, master the differences between several main methods, and some other representations have a general understanding.

2.1 State Space method

Teaching Content: This section solves the problem by the state space method, which is based on the state and the operator (operator) to represent and solve the problem.

Teaching emphasis: The status of the problem description, operator.

Teaching Difficulty: Choose a good state description and state space representation scheme.

Teaching method: Take the classroom teaching as the main, make full use of the multimedia material in the network course to elaborate the abstract concept.

Teaching Requirements: Focus on the state space description of a problem, learn to organize the state space map, use the search graph to solve the problem.

2.1.1 Problem Status Description

1, state of the basic concept

State is an ordered set of least variable q0,q1,...,qn introduced to describe the difference between a class of things, and the vector form is as follows:

Q=[Q0,Q1,..., qn]t (2.1)

Each element Qi (i=0,1,...,n) in the formula is the component of a set, called a state variable. Given a set of values for each component, a specific state is obtained, such as

qk=[q0k,q1k,..., Qnk]t (2.2)

Operator: A means of making a problem change from one state to another is called an operand or a character. An operator can be a walk, process, rule, mathematical operator, operation symbol, or logical symbol.

The state space of the problem is a graph representing all possible states of the problem and their relationships, which contains a collection of three descriptions of all possible problem initial state sets S, operator set F, and Target State set G. Therefore, the state space can be recorded as ternary state (S,F,G).

Question: 1. List the "state" concepts that have been studied and compare them. 2. Enumeration operator.

Example: List several examples of the state and operator in daily life, such as chess.

Discussion: After each step, the chess game has changed, in order to understand the state of the problem space.

2. Representation of State space

For a status description of a problem, 3 things must be determined:

(1) The state description method, especially the initial state description;

(2) the operator set and its effect on the state description;

(3) The characteristics of the target State description.

Examples: Explain the relationship between initial state, operator, intermediate state and target State, and explain the state change process of three digital puzzles.

　

2.1.2 State Diagram method

Basic concepts of graphs

The graph consists of a set of nodes (which are not necessarily finite nodes). A pair of nodes is connected by an arc, pointing from one node to another node. This graph is called a directed graph.

A sequence of nodes (Ni1,ni2,..., nik) When j=2,3,...,k, if there is a successor node NIJ exists for each ni,j-1, then this sequence of nodes is called the path from node ni1 to node Nik length of K.

The cost is to specify the values for each arc to represent the cost of adding to the corresponding operator.

The explicit description of the graph means that each node and its cost curve are clearly given by a table.

The implicit description of graphs means that each node and its cost arc cannot be clearly given by a table.

Question: Take the concepts of "direction diagram", "Path" and "cost" that have been studied.

Examples: Several basic concepts of the state change process for three digital puzzles are explained.

　

2.1.3 State Space Representation example

1, the production of the system

A production system consists of the following 3 parts:

A total database, which contains information about specific tasks.

A set of rules that operate on a database. Each rule consists of a left and a two-part, the applicability or prerequisite of the left-hand identification rule, and the right part describes the action that is performed when the rule is applied. Apply rules to change the database.

A control policy that determines which applicable rule should be applied, and stops the calculation when the database termination condition is met.

2. Example of state space representation

The problem of monkeys and bananas

The state space is represented by a four-tuple (w,x,y,z) in which: the horizontal position of the W-monkey; x when the monkey is on top of the box, take the x=1, or take the horizontal position of the x=0;y-box; z when the monkey picks the banana, take the z=1, or take the z=0.

operator

(1) Goto (U) monkeys walk to the horizontal position U;

(2) Pushbox (v) Monkeys push the box to the horizontal position V;

(3) Climbbox monkeys climbed the top of the box;

(4) Grasp monkeys pick bananas.

The solution process makes the initial state (a,0,b,0). At this point, Goto (U) is the only applicable operation and results in the next state (u,0,b,0). There are now 3 applicable operations, namely Goto (U), Pushbox (V) and Climbbox (if u=b). By continuing to apply all applicable operations to each state, we can get a state space diagram, as shown in the figure. It is not difficult to see from the graph that the sequence of operations that transforms the initial state into the target State is:

{goto (b), Pushbox (c), climbbox,grasp}

For example: The state space diagram of the monkey and banana problem on the multimedia, explain the state space representation of the problem and the application of the production rule.

　

2.2 Problem belongs to Constitution

Teaching content: The Constitution of knowledge representation, that is, the description of known problems, through a series of transformations to turn this problem into a set of child problems, the solution of these sub problems can be directly obtained, thus solving the initial problem of the method.

Teaching emphases: The basic idea of problem reduction, problem description, operator of problem transformation, and or diagram representation.

Teaching difficulties: How to transform the initial problem into a child problem, and or graph representation method.

Teaching Method: Classroom teaching is the main, make full use of the related multimedia material in network course to express abstract concept.

Teaching Request: Through the Vatican tower difficult problem grasps the problem to constitution the mechanism and the question reduction description method emphatically. Learn to use and or graph to represent a problem of reduction.

The description of 2.2.1 problem

1. The concept of problem attribution to Constitution

A description of known problem, through a series of transformations, the problem is eventually turned into a set of child problems, the solution of these sub problems can be directly obtained, thus solving the initial problem.

This method starts with the reverse inference from the target (the problem to be solved), establishes the sub problem and the sub problem of the child problem, until finally, the initial problem is reduced to a common primitive problem set. This is the essence of the problem reduction.

2, the problem belongs to Constitution part

(1) An initial problem description;

(2) A set of operators to transform problems into sub problems;

(3) A set of primitive problem description.

3. Example: The Vatican tower problem

The problem has 3 pillars (1,2,3) and 3 different sizes of discs (a,b,c). There is a hole in the center of each disc, so the disc can be stacked on a pillar. Initially, all 3 discs were stacked on Pillar 1: The largest disc C was at the bottom, and the smallest disc A was at the top. Requires that all discs be moved to column 3, one at a time, and only the disc at the top of the pillar can be moved first, and the larger disc should not be stacked on a smaller disc.

Reduction process

(1) Moving discs A and B to column 2 of the double disk puzzle;

(2) Moving disk C to column 3 of the single disk puzzle;

(3) Moving discs A and B to column 3 double disk puzzle.

From the above can be seen simplifies the problem each is easier than the original problem, so the problem will become easy to solve the primitive problem.

Tell: The source of the Vatican tower problem.

Question: A disk problem to take a few steps. Two disk problems to walk a few steps. Three, four ... Wait

　

4, description of the reduction

The method of problem reduction is to apply the operator to describe the problem as a child problem.

Three-dimensional combinations that can be represented by state space (S, F, G) to specify and describe the problem; for the Vatican tower problem, the sub-question [(111) → (122)],[(122) → (322)] and [(322) → (333)] Specify the foot stone State (122) and ( 322).

The problem reduction method can use the representation of state, operator and target to describe the problem, which does not mean that the problem belongs to constitution and the state space method is the same.

2.2.2 and/or graph representations

1, with or diagram of the concept

A similar graph structure is used to represent the replacement set of the problem to the successor problem, and the graph of the problem is drawn.

For example, imagine that question a needs to be determined by solving questions B, C, and D, and then you can use one and a diagram; Similarly, a problem a, either by solving question B, or by solving problem C, can be represented by one or the graph.

Example: Contains a mixture of diagrams and graphs.

Question: The introduction of additional nodes to a graph or diagram allows each set of subsequent problems to converge under their respective parents ' nodes.

　

2. The relevant terms of the diagram or

The parent node is an initial problem or a problem node that can be decomposed into a child problem;

The child node is an initial problem or a sub problem node of the sub problem decomposition;

Or nodes can solve their parents ' problems as long as they solve a problem;

Nodes with nodes can solve their parents ' problems only if they solve all the problems of children.

The arc is the circular connection of the parent node to the child node;

The end-leaf node is the primitive node corresponding to the original problem.

For example: for one with or diagram.

Question: Indicate the parent node, child node, or node, and node, arc, and end leaf node in the graph.

　

3, and or the relevant definition of the diagram

The general definition of solvable nodes and a solvable node in a graph can be summarized as follows:

(1) The end-leaf nodes are solvable nodes (because they are related to the primitive problem).

(2) If a non-leaf node contains or is a successor node, the non-terminal node is solvable only if at least one of its successor nodes is solvable.

(3) If a non-leaf node contains a successor node, the non-terminal node is solvable as long as its successor node is all solvable.

For example: for one with or diagram.

Question: The end-leaf node, solvable node and irreducible node in the graph are pointed out.

　

The general definition of the irreducible node of an irreducible node is summarized in the following:

(1) Non-terminal leaf nodes with no descendants are not solvable nodes.

(2) If a non-leaf node contains or is a successor node, then the Non-leaf node is not solvable until all its descendants are not solvable.

(3) If a non-leaf node contains a successor node, the non-terminal node is not solvable as long as at least one of its descendants is not solvable.

For example: For the three-disc Vatican tower problem, according to the rules of composition to draw its own map.

Question: The end-leaf node, solvable node and irreducible node in the graph are pointed out.

Homework after class: Textbook chapter Two exercises 2-2 and 2-5

　

4, with or chart composition rules

(1) and or each node in the diagram represents a single problem or set of issues to be resolved. The starting node contained in the figure corresponds to the original problem.

(2) the node corresponding to the primitive problem, called the end-leaf node, has no descendants.

(3) For each possible case where the operator is applied to question A, the problem is transformed into a set of child problems, and a directed arc from a to the successor node represents the set of child problems that are evaluated.

(4) Typically, for each node representing two or more child problem sets, a directed arc points from this node to each node in the set of child issues.

(5) In exceptional cases, when only one operator can be applied to question a, and the operator produces a set with more than one child problem, the graph generated by rule 3 and rule 4 above can be simplified.

2.3 Predicate Logic method

Teaching Content: This section mainly describes the basic method of predicate logical representation of the problem.

Teaching emphases: predicate logic, predicate formula, predicate calculus, permutation and unity.

Teaching difficulties: How to select predicates, the predicate logic representation and operation of the problem.

Teaching Method: Classroom teaching is the main, make full use of the example program in the network course.

Teaching requirements: To grasp the language and method of predicate logic representation, to master predicate formula and predicate calculus, to learn the substitution and unification of predicate formulas, and to use predicate inference to solve problems.

2.3.1 Predicate calculus

1. Syntax and semantics

The basic components of predicate logic are predicate symbols, variable symbols, function symbols, and constant symbols, separated by parentheses, square brackets, curly braces, and commas to denote the relationships within the domain.

An atomic formula consists of a number of predicate symbols and items that have a value T (true) only if the corresponding statement is true within the domain of the definition, and the atomic formula has a value F (false) when its corresponding statement is false within the definition field.

2. Conjunctions and quantifiers

Conjunctions have ∧ (with), ∨ (or), Universal quantifiers (x) and quantifiers (x).

Atomic formula is the basic building block of predicate calculus, and the use of conjunctions can combine multiple atomic formulas to form more complicated and suitable formulas.

3. Several relevant definitions

The formula that is connected by a conjunction ∧ a couple of formulas is called a conjunction, and each part of the combination is called a conjunction. Any combination of the appropriate formulas is also an appropriate formula.

The formula that connects several formulas with the conjunction ∨ is called disjunction, and each part of the disjunction is called an extract. Any disjunction constituted by some suitable formulae is also an appropriate formula.

The formula consisting of two formulas by conjunction → is called implication. The left type of implication is called the preceding paragraph, and the right type is called bottom. If the preceding paragraph and the bottom are all suitable formulae, then implication is also the appropriate formula.

The previous formula with the symbol ~ is called negation. The negation of a suitable formula is also the appropriate formula.

It is also an appropriate formula to quantify the expression of a variable in a suitable formula. If a variable in a suitable formula is quantified, the variable is called a constraint variable, otherwise it is called a free variable. In the appropriate formula, the main interest is that all variables are constrained. Such an appropriate formula is called a sentence.

2.3.2 Predicate formula

1, the definition of the appropriate predicate formula

The recursive definition of the appropriate formula in predicate calculus is as follows:

(1) The atomic predicate formula is the appropriate formula.

(2) If a is the appropriate formula, then ~a is also an appropriate formula.

(3) If both A and B are suitable formulas, then (A∧B), (A∨b), (a=>b) and (A←→B) are all suitable formulae.

(4) If a is the appropriate formula and X is a free variable in a, then (x) A and (x) A are all appropriate formulas.

(5) Only those formulas obtained according to the Rules (1) to (4) above are the appropriate formulas.

For example: Try to represent the following proposition as a predicate formula: any integer or positive or negative.

Question: The quantifiers, conjunctions and implication symbols in the predicate formula of this example are pointed out.

　

2, the nature of the appropriate formula

(1) Negative negation

~ (~p) is equivalent to P

(2) P∨q equivalent to ~p→q

(3) Dick Morgan Law

~ (P∨Q) equivalent to ~p∧~q

~ (P∧Q) equivalent to ~p∨~q

(4) Distribution law

P∧ (Q∨R) is equivalent to (P∧Q) ∨ (p∧r)

P∨ (Q∧R) is equivalent to (P∨Q) ∧ (p∨r)

(5) Exchange law

P∧q equivalent to Q∧p

P∨q equivalent to Q∨p

(6) Binding law

(P∧Q) ∧r equivalent to P∧ (Q∧R)

(P∨Q) ∨r equivalent to P∨ (Q∨R)

(7) Inverse law

P→q equivalent to ~q→~p

In addition, the following equivalence relationships can be established:

(8) ~ (x) P (x) is equivalent to (x) [~p (x)]

~ (x) P (x) is equivalent to (x) [~p (x)]

(9) (x) [P (x) ∧q (x)] is equivalent to

(x) P (x) ∧ (x) Q (x)

(x) [P (x) ∨q (x)] is equivalent to

(x) P (x) ∨ (x) Q (x)

(x) p (x) is equivalent to (y) p (y)

(x) P (x) is equivalent to (y) p (y)

Proof: Negation of negative, ~ (~P) is equivalent to P.

　

2.3.3 Substitution and Oneness

1, replacement

Pseudo-element inference is that the appropriate formula W1 and W1→W2 produce the appropriate formula W2 operation.

The universal inference is that the appropriate formula (x) w (x) produces the appropriate formula W (a), where A is an arbitrary constant symbol.

The permutation of an expression is the substitution of the variable in the expression.

In general, permutations can be combined, but permutations are not interchangeable.

2, Unity

Find the permutation of the item to the variable to make the two expressions consistent, called Oneness (Unification). If a permutation s acts on each element of the expression set {Ei}, the set of the permutation example is represented by {Ei}s. The expression set {Ei} is acceptable. If there is a permutation s make: e1s=e2s=e3s= ... So this is called the oneness of {ei}, because S is the function of making the set {ei} a single form.

For example: the expression p[x,f (y), b] is a permutation of s1={z/x,w/y}, then: P[x,f (y), B]s1=p[z,f (W), b]

　

2.4 Semantic Network method

Teaching Content: This section mainly describes the semantic network representation of knowledge.

Teaching emphases: Lexical, structure, process and semantics of semantic network representation.

Teaching difficulties: How to select nodes and arcs to form the semantic network.

Teaching methods: Classroom teaching.

Teaching Requirements: Focus on the structure of the semantic network, grasp the two-yuan semantic network representation methods to understand the characteristics of the semantic network.

2.4.12 Meta Semantic Network representation 1, the basic concept of semantic network

Semantic network is a structured graphical representation of knowledge, which consists of nodes and arcs or chain lines. Nodes are used to represent entities, concepts, and situations, and arcs are used to represent relationships between nodes.

The semantic network representation consists of the following 4 related parts:

(1) The lexical part determines which symbols are allowed in the glossary, and it involves each node and arc.

(2) The structure part narrates the restriction condition of the symbol arrangement, specifies the node pair that each arc joins.

(3) Part of the process describes the access process, which can be used to create and revise descriptions, and to answer related questions.

(4) The method that the semantic part determines the meaning of the relevant (associative) meanings determines the arrangement of the nodes and their possessions and corresponding arcs.

The semantic network has the following characteristics:

(1) The entity's structure, attributes and the causal relationship between the entity can be expressed explicitly and concisely, and the facts, characteristics and relationships related to the entity can be deduced from the corresponding nodal arcs.

(2) because the concept-related attributes and links are organized in a corresponding node, so that the concept is easy to visit and learn.

(3) Performance problems more intuitive, easier to understand, suitable for knowledge engineers and field experts to communicate.

(4) Semantic interpretation of semantic network structure relies on the inference process of the structure without the contract of structure, so the inference obtained can not guarantee the validity of the predicate logic method.

(5) The relationship between nodes may be linear, tree or network, or even a recursive structure, so that the corresponding knowledge storage and retrieval may require a more complex process. 2, two-yuan semantic network representation

Using two nodes and an arc can represent a simple fact, a semantic network that represents a possessive relationship by allowing a node to represent either an object or a group of objects, or a condition or action. Each condition node can have an outward arc (case arc), called a case box, that describes the various variables associated with the case.

When selecting a node, first make sure that the node is used to represent the basic object or concept, or for a variety of purposes. Otherwise, if the semantic network is used only to represent a particular object or concept, more semantic networks are needed when there are more instances.

The choice of semantic primitives is an attempt to represent knowledge with a set of primitives. These primitives describe the basics and relate to each other in a graphical representation.

For example: The two-yuan semantic network says: XiaoYan is a swallow, the swallow is a bird, the nest-1 is the nest of XiaoYan, and the nest-1 is one of the nests.

2.4.2 representation of multi-meta semantic networks

Semantic network is a kind of network structure. Nodes are linked to each other. In essence, the connection between the contacts is two-yuan relationship. In essence, the semantic network can only represent a two-yuan relationship, if the facts to be expressed are multivariate relations, then the multivariate relationship is transformed into a group of two-ary relationships, or a two-ary relationship. Specifically, multivariate R (X1,X2,...,XN) can always be converted into R1 (x11,x12) ∧r2 (x21,x22) ∧ ... ∧rn (XN1,XN2). Additional nodes need to be introduced to make this transition in the semantic network.

Example: The equivalence between predicate logic and semantic network is illustrated by using the semantic network and predicate logic representation of "Liming is a man".

The expression of 2.4.3 conjunctions and quantification

The semantic network can be used to represent various conjunctions and quantification in predicate logic. 1. Hop-Take

The multivariate relationship can be converted into a set of two-ary relationships, which can be expressed in the form of a semantic network. 2. Extraction

In the semantic network, for the difference of the relation between the conjunction and the connection, the disjunction bounds are raised and the dis is labeled. 3. Negation

Use the ~isa and ~part of the relationship or mark the neg boundary to express the negation. 4. Implication

In the semantic network, the implication relationship can be represented by tagging ante and conse boundaries. 5. Quantification

The existence quantization can be represented directly by the ISA chain in the semantic network. and the full name quantification is to use the segmentation method to express.
2.5 Other methods

Teaching Content: Introduction of the other three representations of knowledge representation, namely, frame representation, script representation and process representation, the principle and application of three representations are expounded.

Teaching emphases: The basic principle and structure of each method.

Teaching Difficulty: The inference process of each method.

Teaching Method: Classroom teaching is the main. Appropriate questions to deepen students understanding of the concept.

Teaching Requirements: Preliminary understanding of the basic principles of three methods.

2.5.1 Framework

1, the structure of the framework

Frames are usually composed of slots that describe the various aspects of things, each of which can have several sides, and each side can have several values. The general structure of a framework is as follows: Frame name

Groove 1〉〈 side 11〉〈 value 111〉 ...

"Side 12〉〈 value 121〉 ..."

 ...

Groove 2〉〈 side 21〉〈 value 211〉 ...

 ...

 ...

Groove N〉〈 side N1〉〈 value n11〉 ...

 ...

"Side Nm〉〈 value nm1〉 ..."

The simpler scenario is to use frames to express things like people and houses. For example, a person can use a description of their occupation, height, and weight, so that these items can be used to form a frame slot. When describing a specific person, then fill in the corresponding slot with the specific values of these items. Table 2.2 shows the framework that describes John.

Table 2.2 Example of a simple framework

JOHN

Isa

:

Person

Profession

:

Programmer

Height

:

1.8m

Weight

:

79kg

　

Framework is a general form of knowledge expression, and there is no uniform form for how to use frame system, which is often decided by different needs of various problems.

2. The reasoning of the framework

As mentioned earlier, a framework is a semantic network of complex structures. Therefore, the matching and attribute inheritance in Semantic network inference can be implemented in the frame system. In addition, because the framework is used to describe things, actions, and events that have fixed forms, it is possible to infer, in new cases, facts that have not been observed. The framework helps to achieve this in several ways: (1) The framework contains information about the situations or objects that it describes.
(2) The frame contains the attributes that the object must have. These properties are used when filling each slot of the frame.
(3) The framework describes the typical examples of the concepts they represent.

The process of using a framework to specifically embody a particular situation is often not very smooth. But when the process encounters obstacles, it often does not have to give up the original effort to start from scratch, but there are many ways to think:

(1) Select the current frame fragment corresponding to the current situation and match the frame fragment to the alternate frame. Select the best match.

(2) Although there is a mismatch between the current framework and the situation to be described, it is still possible to continue to apply the framework.

(3) To query the framework of the special preservation of the chain, to put forward in which direction to test the proposal.

(4) The hierarchical structure arranged along the frame system moves up (i.e. from the dog frame → The mammal frame → the animal frame) until it finds a sufficiently generic framework that does not contradict the existing facts.

2.5.2 Script

A script is a special form of a frame that uses a set of slots to describe the sequence of events, like the sequence of events in a script, which is called a "script" or a script.

A script is generally composed of the following parts:

(1) The opening conditions give a prerequisite for the occurrence of the event described in the script.

(2) The role is used to denote some of the slots of the characters that may appear in the events described in the script.

(3) Props This is used to indicate some of the slots of the objects that may appear in the events described in the script.

(4) The scene describes the actual sequence of events, can be composed of multiple scenes, each scene can be another script.

(5) The result gives the results usually produced after the events described in the script.

Example: Take a restaurant script as an example to illustrate the composition of each part of the script.

According to the importance of the script, there are two ways to prepare the script.

(1) For a script that is not part of the core of the event, simply set a pointer to the script to enable it when it becomes the core.

(2) For scripts that conform to the core of the event, use the specific objects and characters involved in the current event to fill in the script slots. Script prerequisites, props, roles, and events can often play the role of script-enabled indicators.

Once the script is enabled, it can be applied for inference. The most important of these is the use of scripts to predict the occurrence of events that are not clearly mentioned.

Script structure, than the framework of some general structure such as a much more mechanical, the scope of knowledge expression is also very narrow, so it is not suitable for the expression of various knowledge, but for the expression of predefined specific knowledge, such as understanding the storyline, is very effective.

2.5.3 process

Knowledge representation, such as semantic network, framework and script, is a static expression of knowledge and fact, an explicit form of knowledge expression. As for how to use this knowledge, it is decided by control strategy.

and the representation of knowledge is a procedural representation of knowledge. A procedural representation is a process that implicitly expresses the knowledge about a problem area, along with how to use it, as a solution to the problem. What it gives is some objective laws of things, and it expresses how to solve the problem. The description form of knowledge is the procedure, all information is implied in the program. In terms of the efficiency of program solving problems, the programming expression is much higher than the statement expression. But because its knowledge is implied in the program, it is difficult to add new knowledge and expansion function, the scope of application is narrow.

2.6 Summary

There are a lot of knowledge representation methods, this chapter introduces 7 of them, including graphic method and formula method, structured method, declarative representation and procedural representation.

The state space method is a method of problem representation and solution based on the solution space, which is based on the state and operator. When the state space graph is used, the test sequence of the operator is built up from an initial state, each time an operator is added, until the target state is reached. Because the state space method needs to extend too many nodes, it is easy to "combine explosion", so it is only applicable to the problem of simple representation.

The problem is constitution from the target (the problem to be solved), the reverse inference, through a series of transformations, the initial problem is transformed into the set of sub problems and the child problem, until finally it is an ordinary set of primitive problems. The solution of these primitive problems can be directly obtained so as to solve the initial problem, and the solution of the problem of Constitution is effectively explained by the graph. The problem of constitution can express the problem more effectively than the state space method. State space method is a special case of constitution problem. In the Constitution and or graph of the problem, there is a node and or a node, and the state space method contains only or nodes.

The predicate logic method uses predicate appropriate formula and first order predicate calculus to turn the problem to be proved, then uses the digestion theorem and the digestion inversion to prove that a new statement is derived from a known correct statement, which proves that the new statement is also correct. predicate logic is a formal language, which can be symbolic of logical argumentation in mathematics. The predicate logic method is often mixed with other representations, which is flexible and convenient, and can express more complex problems.

A semantic network is a structured representation that consists of nodes and arcs or chain lines. Nodes are used to represent objects, concepts, and states, and arcs are used to represent relationships between nodes. The solution of Semantic network is a new semantic network with definite result, which is obtained by reasoning and matching. Semantic networks can be used to represent multiple relationships, which can be extended to represent more complex problems.

A framework is a structured representation method. A frame is usually made up of slots in various aspects of a specified object, each with several sides, each of which can have several values. Most utility systems must use many frameworks at the same time, and they can be combined into a framework system. Framework representation has been widely used, but not all problems can be expressed in a framework.

A script is a special form of a framework that uses a set of slots to describe the sequence of occurrences. Script representations are especially useful for describing sequential actions or events, but they are not as flexible as frameworks, so the scope of application is not as broad as the framework.

A process is a process representation of knowledge, which implicitly represents a problem-solving process in conjunction with these usage methods. The process representation uses the procedure to describe the problem, has the very high problem solving efficiency. Because knowledge is difficult to operate in the program, the scope of application is narrower.

When expressing and solving more complex problems, it is not enough to adopt a single knowledge representation method. It is often necessary to mix representations in a variety of ways. For example, the comprehensive adoption of framework, semantic network, predicate logic of the process representation (more than two), can make the problem of the study more effective solution.

In addition, when choosing a knowledge representation method, the features and features provided by the programming language used are also considered, so that these representations can be better described.

　

From:http://netclass.csu.edu.cn/jpkc2003/rengongzhineng/rengongzhineng/jiaoan/chapter2.htm