Knowledge graph is a structured representation of facts, which are made up of entities, relations and properties. Entities can be real-world objects or abstract concepts and relationships captures the relations between entities. Knowledge graph and knowledge base is very similar and are often used interchangeably with a minor difference where knowledge graph is a graph representative of the knowledge base and knowledge base is used for formal semantics, interpretations and inference over facts. Below is a figure showcasing the difference between the two:

The KGs Timeline

The concept of semantic net was proposed by Richens back in 1956 while the General Problem Solver was proposed in 1959. Later on around mid 1980s, the community started developing frame-based, rule-based, and hybrid representations. In addition, the Cyc project was started in 1984. Subsequently, Resource Description Framework (RDF) and Web Ontology Language (OWL) were introduced and they became important standards of the Semantic Web in 2001. Since then, many open knowledge bases or ontologies were published! It was, however, only in 2012, where the concept of knowledge graph becomes popular when Google used it for its search engine!

What’s the formal definition of Knowledge Graphs?

As mentioned above, knowledge graph is a structured representation that stores facts. We can define a knowledge graph as G={E, R, F}, where E, R, and F are set of entities, relations and facts. Facts are made of a triplet of (h, r, t), where h, r, and t are head entity, relation, and tail entity. Other definitions of knowledge graphs include:

  1. A knowledge graph acquires and integrates information into an ontology and perform reasoning to derive new knowledge

  2. A knowledge graph is a multi-relational graph that’s made up of entities (nodes) and relations (edges)

Categorisation of Research on Knowledge Graphs

There are four main areas of research on knowledge graphs:

  1. Knowledge Representation Learning (KRL)

  2. Knowledge Acquisition

  3. Temporal Knowledge Graph

  4. Knowledge-aware Applications

Knowledge Representation Learning (KRL)

KRL is important as it’s the foundation for many knowledge acquisition tasks and downstream applications. There are four aspects of KRL:

  1. Representation space – how entities and relations are represented

  2. Scoring function – measures plausibility of factual triples

  3. Encoding models – representing and learning relational interactions

  4. Auxiliary information – additional information to be added into the embedding methods

Knowledge Acquisition

Knowledge acquisition tasks are divided into three categories:

  1. Knowledge Graph Completion

  2. Entity Discovery

  3. Relation Extraction

The first is used to expand existing knowledge graph whereas the last two is used to discover new knowledge from unstructured text.

Temporal Knowledge Graph

Temporal knowledge graphs incorporate temporal information for representation learning. There are four different research fields:

  1. Temporal embedding

  2. Entity dynamics

  3. Temporal relational dependency

  4. Temporal logical reasoning

Knowledge-aware Applications

Lastly, we will look at the knowledge-aware applications, which include natural language understanding, Q&A systems, recommendation systems, and other real-world applications. Overall, see below the diagram for the overview of research on knowledge graphs:



Data Scientist

Leave a Reply