Reasoning over knowledge graphs allows you to discover new knowledge and conclusions from existing data. There are three categories of reasoning methods:

  1. Rule-based

  2. Distribution representation-based

  3. Neural network-based

There are many applications of knowledge graph reasoning. This includes:

  1. Knowledge graph completion

  2. Question-answering

  3. Recommender systems

Reasoning is the foundation of any strong AI agents. The stronger the ability to reason, the better the AI agent performs and reasoning relies on prior knowledge and experience. Knowledge graphs contain huge amount of prior knowledge from different source of data and can also effectively organise the data into a structured, transparent, and easy-to-understand format.

Knowledge reasoning over knowledge graphs can identify errors and discover new relationships between entities and these new relationships can be feed back into the knowledge graph to enrich it.

Introduction to knowledge reasoning

In general, knowledge reasoning is the process of using prior known knowledge to infer new knowledge. Essentially, knowledge reasoning means generating new triplets that previously doesn’t exist in KG. Earlier reasoning works involve two classes:

  1. Logic

  2. Knowledge engineering

The logic class believes that all reasoning was based on first-order logic and predicate logic whereas the knowledge engineering class uses semantic networks to represent concepts and knowledge for describing entities and relations. Both methods rely heavily on expert knowledge.

Introduction of leading knowledge graphs

In 2012, Google introduced its Knowledge Graph (Singhal, 2012). See below the figure of world’s leading KGs.

Essentially, knowledge graphs is a semantic network and a structured semantic knowledge base which can be used to interpret concepts and their relations in the real world.



Data Scientist

Leave a Reply