By Pujun Xie (p-se@edu.hse.ru)

Semantic networks [1] serve as a potent tool within the realm of artificial intelligence (AI), facilitating the representation of knowledge and the comprehension of relationships among diverse concepts. These networks employ graphical depictions, wherein nodes symbolize concepts and edges denote the relationships linking them. Semantic networks find extensive application in natural language processing, the representation of knowledge, and reasoning systems.

Knowledge representation [2] is a branch of artificial intelligence (AI) focused on encoding information about the world in a manner that enables computer systems to tackle intricate tasks. This field integrates insights from psychology on human problem-solving and knowledge representation to devise formalisms that simplify the design and construction of sophisticated systems.

This paper will introduce knowledge representation and semantic networks includes origins, concepts, method, and applications. Besides, we will analyze the relationship between knowledge representation and semantic networks includes the role of semantic networks in knowledge representation.

Knowledge is the knowledge and experience of human beings to the objective world and an information structure formed by relating related information together. The knowledge representation hypothesis [3], first formally defined by Brian C. Smith in 1985, became the official starting point of the field of knowledge representation. Knowledge representation is a general method of machine representation of knowledge, formalizing and modeling human knowledge. It is an area within artificial intelligence dedicated to creating computer-based representations that encapsulate information about the world, enabling the solving of intricate problems. The figure below is an example of knowledge in knowledge representation.

In knowledge representation, methods can be divided into two categories according to their characteristics: declarative knowledge representation and procedural knowledge representation, which represent declarative knowledge and procedural knowledge. Descriptive knowledge representation represents in the form of data separation of knowledge representation from knowledge application concise, rigorous, and inefficient. Procedural knowledge representation is a process combining knowledge representation and knowledge application that is not rigorous enough to be modified easily.

The main knowledge representation methods are following:

1. predicate logic representation
2. semantic networks representation
3. frame representation
4. process representation
5. Petri net representation
6. object-oriented representation
7. Artificial Neural Network (ANN) representation

We will introduce the detail of semantic networks representation in the next section.

Semantic network [3] is a knowledge representation method proposed by Quillian in the 1960s, which is a graphic notation for representing knowledge in patterns of interconnected nodes and arcs. Early versions of semantic networks have long been utilized in philosophy, psychology, and linguistics, whereas computer implementations of semantic networks were initially developed for purposes such as artificial intelligence and machine translation. In semantic networks, interconnected nodes and edges represent knowledge. Nodes represent objects, concepts, and edges represent relationships between nodes. The figure below is an example of semantic network.

Here are the common steps to represent knowledge in semantic networks:

1. Determine objects and their attributes.
2. Determine relationships between objects.
3. Organize nodes (object nodes, action nodes, situation nodes) and arcs according to the relationships involved in the semantic networks.

Semantic network makes it easier for us to understand semantics and semantic relationships. Its expression is simple, straightforward, and in line with nature. However, it lacks of standards for node and edge values; is difficult to integrate multi-source data; is unbale to distinguish between concept nodes and object nodes. So, it is difficult to apply in practice. Subsequent Semantic Web [4] technologies proposed RDF, RDFS, and OWL. They made the semantic web overcome the shortcomings of the semantic networks.

Despite these shortcomings of the semantic networks, there are still many projects built on the idea of the semantic networks. Here are a few of the famous and practical projects:

WordNet: It is an English vocabulary that divides English words into sets of synonyms and associates these sets with different semantic relationships. It has many applications in natural language processing, such as disambiguation, information retrieval, text classification, text summarization, etc.

BabelNet: Compared with WordNet, BabelNet is a multilingual vocabulary. It is built by automatically linking Wikipedia to WordNet, and also uses some other vocabulary resources.

HowNet: It a Chinese semantic dictionary which uses the concepts represented by Chinese and English words as description objects, and builds a common-sense knowledge base that contains the relationship between concepts and the attributes of concepts.

In summary, the relationship between semantic networks and knowledge representation is very obvious. Semantic network is an effective method for knowledge representation, which visually illustrates how concepts are related to each other. As an important role in knowledge representation, the potential of semantic networks still needs to be further explored.

In conclusion, semantic networks play a fundamental role in knowledge representation, offering a unique perspective that facilitates the encoding, understanding, and utilization of knowledge in artificial intelligence systems. As research and technology advance, we can expect further refinements and applications of semantic networks, further cementing their importance in the ever-evolving landscape of AI and knowledge representation.

1. Sowa J F. Semantic networks[J]. Encyclopedia of artificial intelligence, 1992, 2: 1493-1511.

2. Davis R, Shrobe H, Szolovits P. What is a knowledge representation?[J]. AI magazine, 1993, 14(1): 17-17.

3. https://publications.csail.mit.edu/lcs/pubs/pdf/MIT-LCS-TR-272.pdf

4. Davies J, Fensel D, Van Harmelen F. Towards the semantic web[J]. Ontology-driven knowledge management, 2003.