Knowledge map Foundation (I) - what is knowledge map

from ：https://www.jianshu.com/p/cd937f20bf55

Knowledge map basis （ One ）- What is knowledge map

 

Definition of knowledge map

Knowledge map is a relatively new concept in China , At present in China paper Less. , The application side mainly focuses on BAT This kind of enterprise with massive data , The concept is google stay 2012 Put forward in , At that time, it was mainly to integrate the traditional keyword-base The search model is upgraded to semantic based search . Knowledge map can be used to better query complex related information , Understand user intention from the semantic level , Improve search quality .

Personally think that , The biggest advantage of knowledge map is that it has a strong ability to describe data , Although various machine learning algorithms have good prediction ability , But the ability to describe is very weak , The knowledge map just fills this gap .

There are many definitions of knowledge map , Here I provide some of my own understanding ：

1. The main goal of knowledge map is to describe various entities and concepts in the real world , And the strong relationship between them , We use relationship to describe the relationship between two entities , For example, the relationship between Yao Ming and the Rockets , Their attributes , We will use “ attribute -- It's worth it “ To characterize its inherent characteristics , For example, our characters , He has age 、 height 、 Weight attribute .

2. Knowledge maps can be artificially constructed and defined , To describe the weak relationship between various concepts , for example ：“ Forgot the order number ” and “ Retrieve the order number ” The relationship between

The concept of knowledge base

Types of knowledge base

At present, knowledge base can be divided into two types ：Curated KBs and Extracted KBs

Curated KBs： With yago2 and freebase As a representative , They are from Wikipedia and WordNet The knowledge base extracts a large number of entities and entity relationships , It can be understood as a structured Wikipedia .

Extracted KBs： mainly Open Information Extraction (Open IE),  Never-Ending Language Learning (NELL) As a representative , They directly extract entity relation triples from hundreds of millions of web pages . And freebase comparison , The entity knowledge obtained in this way is more diverse , And their entity relations and entities are more in the form of natural language , Such as “ Yao Ming was born in Shanghai .” Can be expressed as (“Yao Ming”, “was also born in”, “Shanghai”). Knowledge extracted directly from web pages , There will also be some noise , Its accuracy is lower than Curated KBs.

At present, it is more commonly used in the industry Curated KBs, Mainly because Curated KBs Relatively simple , Easy to build , Less noise .

What is a knowledge base

a)“ Yao Ming was born in Shanghai ”

b)“ Yao Ming is a basketball player ”

c)“ Yao Ming is the current chairman of the Chinese Basketball Association ”

The above is a piece of knowledge , When a large amount of knowledge is gathered, it becomes a knowledge base （Knowledge Base）. We can wikipedia, Baidu Encyclopedia and other encyclopedias acquire a lot of knowledge . however , The knowledge of these encyclopedias is composed of unstructured natural language , This kind of organization is very suitable for people to read, but it is not suitable for computer processing .

 

chart 1： Example of knowledge map

 

Representation of knowledge base

For the convenience of computer processing and understanding , We need to be more formal 、 To express knowledge in a concise way , That's triple （triple）.

“ Yao Ming was born in Shanghai, China ” It can be expressed as a triple (Yao Ming, PlaceOfBirth, Shanghai)[1]. Here we can simply think of triples as ( Entity entity, Entity relations relation, Entity entity). If we think of entities as nodes , Put the entity relationship （ Including attributes , Categories, etc ） As a side , Then the knowledge base containing a large number of triples becomes a huge knowledge map .

Sometimes entities are called topic, Such as Justin Bieber. There are also two types of entity relations , One is attribute property, One is relationship relation. As shown in the figure below , The biggest difference between attributes and relationships is , Two entities corresponding to the triplet of the attribute , It's often a topic And a string , Such as attribute Type/Gender, The corresponding triplet （Justin Bieber, Type, Person）, And the two entities corresponding to the triple of the relationship , Often two topic. Such as relationship PlaceOfBrith, The corresponding triplet （Justin Bieber, PlaceOfBrith, London）.

 

chart 2：Justin Bieber Knowledge map

（ The blue square in the picture shows topic, The orange ellipse includes attribute values , They all belong to the entities of the knowledge base ; Blue lines indicate relationships , The orange line represents the attribute , They are collectively referred to as the entity relationship of the knowledge base , Can use triples to describe entities and entity relationships ）

Data structure of knowledge base

Here is just a brief introduction to the data structure , Knowledge expression will be in 《 Knowledge map basis （ Two ）- Knowledge representation system of knowledge map 》 In detail .

The reader just needs to remember ,freebase Basic knowledge expression form ：( Entity )-[ Relationship ]-( Entity ),( Entity )-[ Relationship ]-( value ) that will do , Refer to the figure 3, The relationship between Yao Ming and Ye Li .

 

chart 3 Expression of knowledge

The application of knowledge map

Through the map of knowledge , Not only can Internet information be expressed in a form closer to human cognitive world , And it provides a better organization 、 The way to manage and use massive amounts of information . The following figure shows the application of the knowledge map sorted out by the author , In the following articles, the author will analyze the following applications .

 

chart 4 The application of knowledge map

 

From the picture 4 Look up , The application of knowledge map mainly focuses on the field of search and recommendation ,robot（ Customer service robot , Personal assistant ） It's a question and answer system , In essence, it is also an extension of search and recommendation . It may be because of the technology of knowledge map （ especially freebase） It was born to solve the search problem . The knowledge storage part may be that enterprises such as qicha and qixinbao find that the data using graph structure is easier to clean and process .

In semantic search , The search of knowledge map is different from the conventional search , The regular search is based on keyword Find the corresponding web page collection , And then through page rank Wait for the algorithm to rank the web pages in the web page collection , Then show the user ; The search based on knowledge map is to traverse knowledge in the existing map knowledge base , Then return the queried knowledge to the user , Usually if the path is correct , The only knowledge found is 1 One or more , It's quite accurate .

Question and answer system , With the help of knowledge map, the system will first analyze the semantic and grammar of the questions raised by users using natural language , Then it can be transformed into structured query statements , Then search the answers in the knowledge map .