What Is Knowledge Extraction?

Knowledge acquisition refers to the question of how a machine (computer or intelligent machine) acquires knowledge in artificial intelligence and knowledge engineering systems. ^[1]

Knowledge acquisition is a major subject for building knowledge-based systems, but it has not been studied sufficiently. Before the 1960s, most of the knowledge required for artificial intelligence programs was manually programmed into the program by professional programmers. At that time, it was less directly oriented to application systems, and the issue of knowledge acquisition had not yet been given enough attention. With the rise of expert systems and other knowledge-based systems, people have realized that it is necessary to reform backward knowledge acquisition methods, allowing users to build directly and gradually during the operation of the system with the help of knowledge engineers or intelligent programs (knowledge acquisition programs) Required knowledge base.

Including knowledge extraction, knowledge modeling, knowledge conversion, knowledge input, knowledge detection, and reorganization of the knowledge base:

(1) Knowledge extraction: The knowledge contained in the information source is extracted through the processes of identification, understanding, screening, induction, and stored in the knowledge base.

(2) Knowledge modeling: Constructing a knowledge model mainly includes three stages: knowledge identification, knowledge specification, and knowledge refinement.

(3) Knowledge conversion: transform knowledge from one representation to another.

(4) Knowledge storage: The knowledge expressed in an appropriate mode is edited and compiled into the knowledge base.

(5) Knowledge detection: In order to ensure the correctness of the knowledge base, knowledge detection needs to be done well.

(6) Reorganization of the knowledge base: Reorganize the knowledge in the knowledge base to improve the system.

operation efficiency.

Knowledge acquisition manual transfer

Designers, knowledge engineers, programmers, experts, or users who rely on artificial intelligence systems can transplant human knowledge into the machine's knowledge base through system one design, programming, and human-computer interaction or auxiliary tools, so that the machine can acquire knowledge . ^[2]

There are two types of manual transplantation:

(1) Static transplantation

During the system design process, knowledge representation, programming, and establishment of a knowledge base are performed to store, arrange, and manage knowledge so that the system acquires the required prior or static knowledge. It is called "static migration" or "design migration".

(2) Dynamic transplantation

During the operation of the system, through conventional human-computer interaction methods, such as "keyboard-monitor" input / output interaction methods, or auxiliary knowledge acquisition tools, such as knowledge editors, the use of knowledge assimilation and knowledge adaptation technology, and The database is manually added, deleted, modified, supplemented, and updated to enable the system to obtain the required dynamic knowledge, so it is called "dynamic migration" or "running migration".

Knowledge acquisition machine learning

During the operation of the artificial intelligence system, it learns, acquires knowledge, accumulates knowledge, and adds, deletes, modifies, expands, and updates the knowledge base.

There are two ways of machine learning:

(1) Teaching-based learning

In the process of machine learning, human beings serve as instructors or supervisors, give evaluation criteria or judgment standards, check the working effect of a system, select or control the "training set", and guide and supervise the learning process. This learning method is usually offline, non-real-time learning, but also online and real-time learning.

(2) Self-study learning

In the process of machine learning, there is no need for a person to be a teacher or supervisor, but the supervisor of the system itself implements the supervision function to carry out the learning process. Supervision, provide evaluation criteria and judgment standards, and check the work effect through feedback, control Selection and training. This learning method is usually online, real-time learning.

Knowledge acquisition machine perception

In the process of debugging or running, the artificial intelligence system directly senses the external world through machine vision, machine hearing, and machine touch, etc., inputs natural information, and acquires perceptual and rational knowledge.

There are two main ways of machine perception:

(1) Machine vision

During system debugging or operation, machine vision such as text recognition, image recognition, and scene analysis is used to directly input the natural information of the corresponding text, images, and scenes from the external world to obtain perceptual knowledge. After recognition, analysis, and Understand and gain relevant knowledge.

(2) Machine hearing

In the process of system debugging or running, through machine hearing such as voice recognition, language recognition, and language understanding, directly input the corresponding sound, language and other natural information from the outside world to obtain perceptual knowledge, and through recognition, analysis and understanding, obtain relevant rationality Knowledge.