An introduction to knowledge graphs

In today’s world, data is everywhere—and it’s growing at an unprecedented pace. But having data isn’t enough. The real challenge lies in making sense of it all. How do we move beyond simply collecting isolated facts to uncovering the meaningful connections between them? Enter knowledge graphs - the key to translating raw data into key insights!

A knowledge graph is a sophisticated, interconnected system for organizing information. Rather than storing isolated facts, it maps how each key piece of information is associated with  others. For example, a knowledge graph will display how a person or a product (called a “node”) is connected to a location availability (labeled an “edge”). Knowledge graphs can also be enhanced with metadata if needed, e.g., to track the provenance (origin) of the data, or the relative strength / certainty of the asserted relationship. This web of relationships allows for a more contextual  understanding of data.

Knowledge graphs are increasingly vital tools, and at CANA, we apply them across a range of projects - such as mapping global supply chains. These maps allow us to explore intricate details, from the sourcing of individual parts, the components that might be within a part, locations of manufacturers, political and legal environments in which they operate, and even corporate leadership structures. By connecting all this  information, we’re able to analyze the broader context and extract meaningful, actionable insights. We use knowledge graphs to understand not just isolated data points, but the relationships between them - because, as we all know, businesses do not operate within vacuums. What happens around them is just as critical as what happens within. 

So what are the big benefits of knowledge graphs?

They Provide Context: A knowledge graph doesn't just tell you "Widgets R Us is a company." It can also tell you "Widgets R Us is a company that makes widgets," and "Widgets are a type of antenna," and "antennas are transducers." This helps us understand the bigger picture.

They Enable Smarter Search and Discovery: Users can ask more complex, conceptual questions. For example, CANA could easily find "all the suppliers of Meals-Ready-To-Eat" or "the number of American companies that build commercial drones." The graph understands the underlying meaning of your query.

They Power AI and Machine Learning: The organized, clearly defined structure of knowledge graphs is a boon for AI and its ability to learn and reason from information. This makes AI applications more accurate, insightful, and capable of performing more complex tasks.

They Facilitate Data Integration: Knowledge graphs can eliminate the siloing of information. They act as unifiers, connecting disparate data sources and creating an enterprise-wide view of information.

At its core, a knowledge graph often uses a structure called a "triple" to represent information. A triple consists of a Subject (an entity, e.g., "Brand X Trucking"), a Predicate or Relationship (how the subject relates to the object, e.g., "is a partner of"), and an Object (the entity the subject is related to, e.g., "X Mining Corporation"). A triple can represent an internal property of a node (e.g., Brand X trucking was founded in Y year) or a connection between nodes. So, a simple piece of information like "Brand X trucking is a partner of X Mining Corporation" becomes a clear, interconnected part of the graph, ready to be linked with other facts. In this example, the “triple” structure is considered an atomic unit - the smallest meaningful piece of information the knowledge graph can use.

Building a Knowledge Graph

So you’re ready to build a knowledge graph, how do you actually do it? It’s a fascinating blend of data engineering and cutting-edge artificial intelligence (AI). If you’re curious to see one in action without all the heavy lifting, you can leverage a generative AI tool. Simply ask it to generate the code for a knowledge graph on a topic of your choice. For instance, prompting an AI tool with “write the code for a knowledge graph that looks at drones in the United States since 2000” will produce an HTML application.

This application, often powered by D3.js, visually represents a force-directed graph, showcasing the intricate  relationships between different entities like drones, manufacturers, and regulations. You can even enhance the detail by requesting for association scores, and the tool will incorporate weighted connections to show the strength of these relationships. The real beauty here is the ease with which  anyone can copy, paste, and adapt this generated code to suit their specific needs. 

This provides an informative, albeit basic, starting point.  For more robust applications, significant data validation and additional data would be necessary. However, these skills aren’t just for serious data analysis; they can be applied to other engaging and entertaining purposes as well. 

Imagine, for example, requesting a graph that visualizes the number of books about zombies by the author’s country of origin. The result, much like the previous example, would be generated almost instantaneously!

If your interest is piqued, what are some things to consider in building a graph on your own? 

Ontology Design: First, the graph's purpose is defined and then its "rules of the road”. An ontology lays out all the types of entities in your graph - like "Person" or "Company" - and defines the properties these entities can have. It also specifies how relationships between entities are defined, such as "works for" or "manufactures". The ontology can be used for data validation, too, which is helpful for AI agents. For example, if an AI agent generated a knowledge graph from a news article, and some of its triples didn't abide by the ontology's vocabulary, you can pass the erroneous triples back to the agent asking it to correct the triples before proceeding. Think of it as a comprehensive rulebook, roster, and referee rolled into one.  

Data Collection & Preprocessing: Data for your knowledge graph can come from everywhere – databases, files, and especially unstructured text from sources like webpages and documents. This could be anything from a few documents or enormous sets of what is aptly called big data. Just like spring cleaning, once you pull everything out of the metaphorical drawers and closets, it all needs to be sorted, cleaned, and organized. Advanced AI like Natural Language Processing (NLP) and Large Language Models (LLMs) are crucial here. They can be used to identify entities (like people, places, and things) within text and determine the relationships between them. In turn, duplicate items are removed, conflicts are resolved, and everything is standardized. There’s no hoarding allowed! 

Data Ingestion & Storage: Knowledge graphs live in special graph databases (like Neo4j or Amazon Neptune). These offer storage and the ability to quickly navigate the intricate web of nodes and edges.

Refinement & Maintenance: A knowledge graph is a tool that constantly evolves! It can deduce new facts that weren't explicitly added, like the transitive property we learned in high school. For example, if it knows "A is a type of B" and "B is a type of C," it can deduce "A is a type of C." What’s more, when new information comes in, the graph is continuously updated, ensuring it remains relevant and accurate.

This overview barely scratches the surface of the power and complexity of knowledge graphs. While numerous tools are available to get you started, be prepared for challenges; even the most well-designed graphs require ongoing refinement. Stay tuned for Part II of An Introduction to Knowledge Graphs where we’ll delve into some of the specific challenges the CANA team has encountered over time and the solutions we’ve developed.