Advantages:
1. Representation is natural, easy to understand, and widely used;
2. In line with human associative memory;
3. Structured knowledge representation.
Disadvantages :
1. Not strict: there is no recognized logical basis;
2. It is difficult to handle effectively: the network form has the characteristics of flexibility, but
at the same time it causes the low efficiency of processing and retrieval.

2.2 Production system

Advantages:
1. Naturalness: It is in line with the knowledge representation form of human expression of causality, which is intuitive and natural, and is convenient for reasoning.
2. Modularity: The rules in the production system have the same form, which is easy to modularize management.
3. Effectiveness: It can express deterministic knowledge, uncertain knowledge, enlightening knowledge, process knowledge, etc.
4. Clarity: The format is fixed, which is convenient for rule design, and it is easy to check consistency and integrity in the rule library. .
Inadequate:
1. Efficiency is not high: high cost of matching rules, to solve complex problems easily lead to combinatorial explosion.
2. Cannot express structural knowledge: Cannot express the differences and connections between things with structural relationships

2.3 Frame Systems

Frame: The frame is the basic unit of knowledge representation, a data structure that describes the attributes of an object (things, events, concepts, etc.). A frame
is composed of thousands of " slots" (Slot) structure, each slot can be divided into several " face" (Facet)
slots: describe a certain aspect of the attribute
side: describe an aspect of the corresponding attribute, usually An attribute value

Advantages
1. The description of knowledge by the framework is very complete and comprehensive;
2. The quality of the knowledge base based on the framework is very high;
3. The framework allows numerical calculation and is superior to other representation languages at the time;
less than
1. The construction cost of the framework is very high, and the quality requirements of the knowledge base are very high
;
2. The expression of the framework is not flexible, and it is difficult to use it in conjunction with other forms of data . (Data Island)

2.4 Conceptual Graph

2.4.1 Description Logic

Horn clause

2.4.2 Description logic

Composition of description logic

Concept (Concept): describe abstract terms in the world of
relationships (Role): the link between concept
instances (Individual): The only individual
axioms (Axiom): self-evident proposition
, for example,

I did not understand the reasoning of the description logic ALC

3.1 RDF(Resource Description Framework)

aim of design:

Minimal constraints, flexible information per second, can be used on the Web

1. A graph model based on triple declarations (graph representation)
2. A URI-based extensible vocabulary set (unique identification)
3. XML-based serialization syntax encoding (information exchange)
4. Formal semantics and availability Proof of inference (logic basis)
5. Allow anyone to make a statement of any resource (open world)

3.1.1 RDF model

Resources (Resource): everything URI identifies
text (Literal): or a string data type value
property (Property): Description resource characteristics, attributes, or relationship
declaration (Statement): plus a resource attributes and attribute values.

3.1.2 RDF Schema (Diagram)

RDFS is used to define and describe vocabulary sets

Class: rdfs:Class
Class hierarchy: rdfs:subClassOf
Instance definition: rdfs:type
Property definition: rdfs:range, rdfs:domain
Property hierarchy: rdfs:subPropertyOf

3.1.3 Specific examples

RDF graph about a book

About RDF language to describe the Person class

3.1.4 Insufficiency of RDF
1. Definition of the value range : RDF(S) defines the value range of the attribute through rdfs:range. The value range is global and cannot be explained when the attribute is applied to some specific classes. The special range limit of.
2. Equivalence of classes, attributes, and individuals : RDF(S) cannot declare whether two or more classes, attributes, and individuals are equivalent or unequal.
3. Definition of disjoint classes : In RDF(S), only subclass relations can be declared. For example, both men and women are subclasses of human beings, but they cannot be declared as disjoint.
4. The definition of Boolean combination of classes : that is, the combination of certain classes is realized through the declaration of class union, intercourse and complement, thereby constructing new classes, such as defining humans as the union of the two classes of men and women.
5. Cardinality constraint : that is, to restrict the possible or necessary value range of a certain attribute value , such as stating that one person has parents (including two people), and there is at least one teacher in a course.
6. Description of attribute characteristics : that is, to declare certain attributes of attributes, such as transitivity, functionality, symmetry, and to declare that one attribute is the inverse attribute of another attribute.

3.2 OWL （Web Ontology Language）

3.2,1 OWL design ideas

1. Extend RDFS, use RDFS/XML in grammar rules, and strictly follow description logic in semantics.
2. According to the application's different requirements for expressive ability and reasoning complexity, OWL provides two sub-languages, OWL DL and OWL Lite.
3. In order to be fully compatible with RDFS, OWL Full is provided, which includes all the contents of OWLDL, but this also causes the OWL Full reasoning problem
to be undecidable.

3.3 SPARQL

Based on the graph matching model, the SELECT-FROM-WHERE sentence is used, and the graph operator OPTIONAL is enhanced

3.4 JSON-LD(JSON for Linked Data)

4 Knowledge Graph Statistical Representation Learning ( Representation Learning )

Traditional representation:
symbol-based triple representation, which cannot capture the semantic relationship between entities (explicit + implicit). It is
not good to directly use various machine learning models for analysis and mining.
Statistical learning representation:
give a statistical distribution Expression form,
can capture the semantic relationship between entities, especially the implicit relationship

The representation is a vector, which
can be directly used by various machine learning models.

4.1 Representation learning in natural language

4.2 Knowledge Graph Represents Learning Principle

Reference materials:

Introduction to Knowledge Graph (2) Knowledge Representation and Knowledge Modeling https://blog.csdn.net/pelhans/article/details/80020067
Southeast University PPT

Knowledge Graph Introduction Study Notes (2)-Knowledge Representation

1 The concept of knowledge representation

1.1 Knowledge representation method

1.2 The classification view of knowledge representation

1.3 The development of knowledge representation

2.1 Semantic Network