Domain-specific knowledge graph examples
Example 1: Academic Publication Domain
Below is the figure of an academic publication knowledge graph. The two purple nodes represents the different publications under their titles. The two publications have other connections such as authors, dates of publication and venues. The figure showcase the expressiveness in representation. There’s a distinction between entities and literals. Entities can have relationships with other entities or attributes. In the triplets, the tail connection could either be an attribute or an entity. The head connection must always be an entity.
Example 2: Products and Companies
Here’s another domain-specific knowledge graph in the products and e-commerce domain. As illustrated below, we can see the distinction between connections of entities and attributes. As you can see, there are numerous degrees of freedom when modelling the same product, being a generic iphone as an entity or a specific iphone model. This choice of modelling can impact both the upstream and downstream tasks such as entity resolution and querying! In addition, not all entities have equal amount of information and aggregation between entities are not always trivial. It’s a common problem that the same entity is extracted independently from multiple raw sources, which means that we have to perform entity resolution on the extracted KG to ensure that there are no duplicated entities.
Example 3: Geopolitical Events
This is by far the most complex example of a geopolitical event knowledge graph. Events are represented as “second-order” entities because they tend to use “first-order” entities such as locations and times as their arguments, which in turn these “first-order” entities have their own attributes. In addition, events can also have attributes of their own and too has relationships with other entities. Extracting and resolving events is another area of research in its own term and performance on them are poor relatively to extracting and resolving “first-order” entities. Check out GDELT for a good example of an event knowledge graph.