Knowledge atlas: knowledge representation

0. introduction

In the map of knowledge , The structured representation of knowledge mainly includes Symbolic representation and Vector representation Two kinds of methods . In the early , First order predicate logic is often used （First Order Logic）、 Semantic Web （semantic network）、 Description logic （Description Logic） And framework system （Frame System） And other knowledge representation methods based on symbolic logic . and At present , It mainly uses the form of triples based on graph data structure （ Head entity , Relationship , Tailstock ） To symbolically represent knowledge .

1. Symbolic knowledge representation based on graph

Symbolic knowledge representation based on graph , The most commonly used are attribute graphs （Property Graph） and RDF chart （Resource Description Framework, Resource Description Framework ） Two kinds of .

• Attribute map （Property Graph） It is the most commonly used knowledge map modeling method in industry , Its advantage is that it allows for entities （entity） Or relationship （relation） Add attribute （property¹）, The expression is very flexible 、 It is very convenient to express multiple relationships , And the storage of attribute graph （ Use graph database Neo4j） Make full use of the characteristics of graph structure to optimize the performance , It has great advantages in query calculation ; However, the deficiency is that attribute maps lack the support of Industrial Standards , And it can't support symbolic logic reasoning .

• RDF（Resource Description Framework², Resource Description Framework ） yes W3C（ International World Wide Web Consortium ） Promote semantic data exchange standards and specifications ³, Its basic constituent unit is triple （ namely ：( Head entity , Relationship , Tailstock )）, It has a more rigorous semantic logic basis , Support logical reasoning , And it is more compatible with semantic representation （ And more complicated ） Ontology representation language OWL.

• need Additional explanation Yes. ： In the actual ,RDF The figure is using RDFS The establishment of a .RDF Classes and attributes are used to describe the relationship between entities , These classes and attributes are defined by the schema （Schema） To define .RDFS（Resource Description Framework Schema,RDF Pattern ） Is in RDF On this basis, it provides a simple description of classes and attributes , So as to give RDF Data provides a standard vocabulary modeling language .

• OWL（Ontology Wed Language, Network ontology language ） As a complete ontology representation language , When triples cannot meet the needs of semantic representation , It provides more alternative semantic expression components .

2. Vectorized knowledge representation

After completing the symbolic representation of graph based knowledge , You can also learn the characteristics of graphs , Get a vector representation for each entity and relationship in the graph ; And use vectors 、 Calculation between matrices or Tensors , Achieve efficient （ knowledge ） Reasoned calculation .

Vectorial representation of knowledge , It is a research hotspot at present .

---

1. attribute （property） It refers to the definition of entity or relationship 、 shape 、 Descriptive information such as complements . ︎

2. Resource： page 、 picture 、 Video and anything else with URL Identifier resource ;
   Description： attribute 、 The relationship between features and resources ;
   Framework： Model 、 Language and the syntax of these descriptions . ︎

3. Because in the early days Web（ namely ：Web 2.0） The standard language of HTML and XML Unable to meet the requirements of semantic web for knowledge representation , therefore W3C Propose a new standard language RDF、RDFS and OWL, To represent the knowledge system of Semantic Web . ︎