This repository contains example code for building knowledge graphs using the below graph databases and their client libraries.
- Neo4j
- Grakn
A knowledge graph uses an ontology to bind connected data into a knowledge domain that can then be reasoned with programmatically. It makes use of specialized data structures defined within its underlying graph database to efficiently store and query connected data.
Simply put, knowledge graphs allow us to represent highly connected data in a more natural way. By representing entities and how they are related to one another in the real world, we can ask intuitive questions such as "How does person X know person Y, and how many common interests do they share?". Since data is stored natively in a connected format, it becomes very efficient to traverse the paths between connected entities in the graph to answer these questions, regardless of the size of the graph.
The following datasets are used as case studies for each different graph database.
phone_calls- Dataset of persons and their phone calls, obtained from the Grakn documentation examples.social_network- Dataset of a simple, artificial social network that connects people and the places they live in.
The world's most popular graph database, Neo4j uses a Labelled Property Graph model to store data natively as a graph. This is done by defining "nodes" that represent an entity in the real world (such as a person or a company) and "edges" that represent how these nodes are connected to one another via triples - for example, person-A -[WORKS_AT]- company-B. The primary benefit of using a property graph model is that it is intuitive and easy to understand - it is very simple to get a Neo4j graph model up and running.
Strictly speaking, Grakn is a "true" knowledge graph, in the sense that it models data using an entity-relationship model that makes use of multiple inheritance of hierarchies, hyper-entities and hyper-relations - this is referred to as a "hypergraph model".
A key feature of hypergraphs is that they generalize the notion of edges (from graph theory) to be composed of a collection of nodes (each of which represents a relationship). This allows not only entities to be related to one another, but also relationships to be related to other relationships. In addition, thanks to its ontology layer and schema definitions, Grakn can perform automated reasoning by utilizing user-defined rules to identify inferred relationships that were not explicitly defined in the graph structure.
The installation instructions for Neo4j on Mac/Ubuntu/Windows are shown on their web page.
The installation instructions for Grakn on Mac/Ubuntu/Windows are shown on their web page.
To more easily work with the raw data while building the graph, we use the Python clients for Neo4j and Grakn.
It is recommended to set up a virtual environment and install the required Python clients using requirements.txt as follows:
python3 -m venv venv
source venv/bin/activate
pip3 install -r requirements.txt
For further development, simply activate the existing virtual environment.
source venv/bin/activate