Executive summary:
A knowledge graph is a defined ontology expressed in computer code. It represents information as a network of entities and their relationships. By structuring data in this way, knowledge graphs enable machines to interpret complex information, mirroring how humans make connections between concepts.
Knowledge graphs can be layered with generative AI such as large language models to create neurosymbolic ensembles. Neurosymbolic AI is an exciting new field that leverages the flexibility and creativity of generative AI with the groundedness of symbolic AI (e.g., knowledge graphs).
Talbot West has pioneered the combination of diverse AI/ML capabilities for a variety of ensembles, including neurosymbolic capabilities. If you’re looking to enhance your organization's capabilities, contact us for a free consultation to explore how we can assist with your digital transformation goals.
A knowledge graph helps machines understand and retrieve information by structuring data in a way that mimics human logic.
A knowledge graph is a structured network of facts. It organizes information as a web of entities (nodes) and their relationships (edges). Nodes represent people, places, or objects, while edges define how nodes connect. This structure allows humans and machines alike to easily explore relationships between named entities.
Each node represents a fact, and each fact is usually stored as a triple (subject-predicate-object), such as [Apple Inc. | is headquartered in | Cupertino].
Knowledge graphs enable richer, semantic understanding for tasks like search, recommendation, and AI reasoning.
Knowledge graphs rely on semantic relationships to model data. This means they use context, meaning, and relationships between entities to organize complex and diverse information. Here's how:
In the past, knowledge graphs were primarily used for data integration and to power semantic search engines such as Google.
Knowledge graphs helped organize and link information by showing how concepts and facts relate to each other. This made it easier for systems to pull together data from lots of places without just matching keywords, giving users more useful and direct answers.
Knowledge graphs help search engines understand what users are really looking for by anchoring search terms to the different entities associated with those terms. The knowledge graph maps connections between people, places, things, and ideas, and the search engine tries to match a user's query with the corresponding entity the user had in mind.
Neurosymbolic AI is a new kind of artificial intelligence that blends two proven methods:
This combination lets AI systems both “learn” from examples (such as recognizing faces or understanding text) and “reason” by applying business logic or rule (like checking compliance or explaining why a recommendation was made). The big benefit for business is that neurosymbolic AI can make smarter decisions with fewer mistakes, offer transparency about how it works, and handle complex challenges that require both pattern recognition and common-sense reasoning.
One category of neurosymbolic AI system is that of a knowledge graph paired with a deep neural network, for example a large language model. This combination gives the combined capability to "know" facts grounded in the knowledge graph as well as "learn" patterns from data.
Launched in 2012, Google’s Knowledge Graph brought the concept into the mainstream by enhancing search with a structured database of interconnected facts about people, places, and things.
While Google has not released many technical details about its exact structure or size, the graph is central to powering features such as Google search suggestions and rich snippets. External access to this knowledge graph is limited, as it’s primarily used within Google’s ecosystem.
DBpedia extracts structured data from Wikipedia’s infoboxes to create a vast knowledge graph with over 4.5 million entities. It spans many topics, including people, places, organizations, and species.
As a key project in the Open Linked Data movement, DBpedia has become a valuable resource for building internal knowledge graphs by providing access to crowdsourced, encyclopedic data. Its comprehensive ontology helps organizations enrich their data with globally recognized entities.
GeoNames offers a rich dataset of over 25 million geographical entities and features and is one of the most comprehensive geographic knowledge graphs available.
Accessible under a Creative Commons license, GeoNames provides detailed information on countries, cities, landmarks, and other geographical features, supporting applications such as mapping, logistics, and geographic data analysis.
WordNet is a large lexical database for the English language that groups words into sets of synonyms and provides definitions, examples, and their relationships.
This knowledge graph is commonly used in natural language processing (NLP) and search applications to improve text understanding and boost query accuracy by supplying rich semantic connections between words.
Developed by Ontotext, FactForge aggregates Linked Open Data from different sources, including DBpedia, GeoNames, and specialized ontologies such as the Financial Industry Business Ontology.
It integrates these data sets with news articles and is a powerful resource for organizations that need real-time information about people, organizations, and locations. FactForge is widely used in industries such as finance and publishing to power knowledge-driven applications.
Cognitive hive AI (CHAI) can use knowledge graphs to shape a dynamic, adaptable ontology for data-driven decisions. It then pairs those knowledge graphs with an ensemble of other types of AI and machine learning. The knowledge graph grounds other parts of the CHAI system and keeps the system more explainable. This neurosymbolic CHAI approach has applications across many industries:
CHAI's modular, composable AI structure makes it infinitely configurable, especially to the specific neurosymbolic capabilities that a given application demands.
At Talbot West, we specialize in implementing CHAI’s adaptable AI solutions. Whether you're exploring AI for your business or need a feasibility study, pilot project, or tool assessment, we’re here to help. Schedule a free consultation to discover how CHAI can meet your strategic needs, and visit our services page for the full scope of our offerings.
The largest knowledge graphs are Google's Knowledge Graph, Microsoft's LinkedIn Graph, and Facebook’s Entity Graph. These are vast graph databases that cover billions of real-world entities, relationships, and concepts from public sources, continuously updated to stay current.
Google Search uses a knowledge graph but isn't one itself. The graph supports search functions by providing structured information about entities and their relationships. This data appears in knowledge panels and enriches search query results with contextual information.
A knowledge graph represents a limited application of an ontology. A knowledge graph represents data with entities and relationships, while an ontology defines the formal semantics—rules, classes, and structures—underpinning that graph. Ontologies often guide the construction of semantic networks.
Knowledge graphs and vector databases are different. Knowledge graphs represent relationships between entities, while vector databases store vectors as mathematical relationships in a high-dimensional space. Knowledge graphs focus on semantic representations of data, whereas vector databases excel at similarity searches and semantic search capabilities.
Large language models typically don't use knowledge graphs directly. They learn from text data during training. With the CHAI paradigm, a knowledge graph and an LLM (or multiple) can interact and dynamically inform one another.
Node embeddings transform nodes in a graph into dense vector representations. These vectors capture the structural and semantic properties of nodes in a lower-dimensional space. Knowledge graph embeddings facilitate tasks such as link prediction, node classification, and integration with machine learning models.
Talbot West provides digital transformation strategy and AI implementation solutions to enterprise, mid-market, and public-sector organizations. From prioritization and roadmapping through deployment and training, we own the entire digital transformation lifecycle. Our leaders have decades of enterprise experience in big data, machine learning, and AI technologies, and we're acclaimed for our human-first element.
The Applied AI Podcast focuses on value creation with AI technologies. Hosted by Talbot West CEO Jacob Andra, it brings in-the-trenches insights from AI practitioners. Watch on YouTube and find it on Apple Podcasts, Spotify, and other streaming services.
