How to build a Knowledge Graph?

step 1

Start Small and Grow

Your efforts to implement these technologies will probably have to compete with other initiatives for the resources and funds. Start by building a solid business case for knowledge graphs and semantic AI. To do that, select a small and concrete use case that shows the business value a knowledge graph can bring to your organization. This will help you gain support and buy-in.

Remember that effective business use cases are driven by strategic goals. Clearly define the business value of your use case by explaining how it makes processes or services more efficient and intelligent for the enterprise.

Some of the most relevant use cases for implementing knowledge graphs and AI are:

• Knowledge discovery: intuitive search and analytics using natural language
• Recommender systems: discovering related data and content
• Semantic data catalogs: agile data integration
• Semantic AI: compliance and operational risk prediction
• Customer 360: unified views of customers and personalization

step 2

Get To Know Your Data

The next thing you need to do is gain a good overview of your data landscape. There are different approaches for inventorying and organizing enterprise data. To determine which types of content are relevant to your use case, consult with subject matter experts and analyze your data. If you are faced with a large number of items, there are automation tools that can help you.

Most companies work with large amounts of unstructured data, such as emails, reports, presentations and other text files. Taxonomies help to classify content and to organize your data and are the starting point for a data catalog! When based on machine-readable standards like SKOS, taxonomies also lay the foundation for even richer semantic models such as ontologies to automate data integration. A business taxonomy provides structure to otherwise unstructured information. It can help you capture, manage, and derive meaning from large amounts of data and content.

When selecting data for your prototype, make sure that it:

• contains both structured and unstructured data so you learn to work with both,
• is not too volatile so you do not have to deal with synchronization at the beginning,
• is not too big so you do not have to deal with performance at the beginning,
• choose data sources that when connected can do/show something that was not possible before.

step 3

Form a Working Team

A precise and detailed view of the roles involved such as taxonomists will also help to define appropriate skills and tasks to bridge mental differences between departments, which focus on data-driven practices on the one hand, and more on documents and knowledge-based work on the other.

Similarly, the question of how subject matter experts with strong domain knowledge (and possibly little technical understanding) can work together with data engineers who are able to use strongly ontology-driven approaches to automate data processes as efficiently as possible is also addressed.

Also involving business users and ‘citizen data scientists’ as soon as possible is essential, since users will become an integral part of the continuous knowledge graph development process nurturing the graph with change requests and suggestions for improvement.

step 4

Map Relationships Across Your Data

Once you have a well-defined prototype and know exactly what data you want to use, it is time for your team to start creating taxonomies and ontologies. But before you start, see what is already available. There are many well-developed taxonomies and ontologies out there for different domains, commercial and non-commercial. Do not start building something from scratch before evaluating if there is something out there you can reuse.

With the help of ontologies, connections between information and data from different sources can be created automatically. Ontologies enable you to map relationships between concepts in a single location at varying levels of detail. This sets the groundwork for intelligent AI capabilities, such as text mining and context-based recommendations. Here are some other things you can do with ontologies:

• Reuse hidden and unknown information
• Manage content more efficiently
• Optimize metadata management to improve search
• Create relationships between disparate and distributed data

Taxonomies and ontologies are a powerful method to map the actual business logic to all existing data models without having to significantly change the existing data landscape. This allows you to link your domain knowledge with your data in an agile way and analyze it as a whole. Ontologies also support the ongoing development of the knowledge graph, as they can be used to perform automatic data quality and consistency checks.

step 5

Stay Agile

Most likely you will be successful with your first pilot application built on graphs. You have excited several stakeholders in your company, and even non-technical people have quickly grasped the beauty of graph technologies. People from other departments start asking what is in for them.

Now you are in a critical phase, as you may want to try to make the big change and plan it for the next 20 years. Don’t do that! Experiment in order to make valid decisions based on experience. Learn that experiments are not bad things or even a sign of immaturity, but rather the only chance to learn, to become better, to improve continuously and to develop skills. Agile is everywhere these days.

We know that, but we also need agile access to data to make better use of it. So we need agile data management. A knowledge graph project must always be an agile data management project. Knowledge is a living thing that is constantly changing.

“Change is the only constant in life.” (Heraclitus of Ephesus)