← Fintech GlossaryAI & LLM

Knowledge Graph

A knowledge graph is a structured representation of knowledge that organizes information as entities (nodes) and their relationships (edges), creating a network of interconnected concepts. Unlike traditional databases that store data in tables, knowledge graphs store information in a graph structure that captures the semantic relationships between entities, enabling sophisticated reasoning, inference, and traversal of connected information. Knowledge graphs are built by extracting entities and relationships from structured and unstructured data sources, then organizing them into a graph model using ontologies or schemas that define the types of entities and relationships. Each entity in the graph has properties or attributes that describe its characteristics, and each relationship has a type that describes how the entities are connected. For example, in a financial knowledge graph, entities might include companies, executives, financial instruments, regulatory filings, and market events, with relationships like 'CEO_of,' 'issued,' 'filed,' and 'impacted_by.' Knowledge graphs support a wide range of AI capabilities including semantic search, where queries can traverse the graph to find related entities; reasoning, where the graph can infer new relationships based on existing ones; and explainability, where the graph provides a transparent, traceable path from a question to an answer. The construction of a knowledge graph involves several stages: entity extraction from source documents, entity resolution to identify when different mentions refer to the same real-world entity, relationship extraction to identify connections between entities, and graph construction to organize everything into a coherent structure. Knowledge graphs can be built manually by domain experts, automatically through NLP and machine learning, or through a combination of both approaches. Modern knowledge graphs in financial services often incorporate LLM-based extraction and reasoning, using LLMs to identify entities and relationships from unstructured text and to answer questions by traversing the graph. The combination of knowledge graphs with RAG systems is a powerful architecture that provides structured factual grounding for LLM responses.

In Financial Services

In financial services, knowledge graphs are used for a wide range of applications including financial crime detection, regulatory compliance, investment research, and risk management. One of the most common applications is in anti-money laundering and financial crime detection, where knowledge graphs model the relationships between accounts, transactions, individuals, and entities. By traversing the graph, investigators can identify suspicious patterns such as circular transactions, layered ownership structures, and hidden beneficial ownership that would be difficult to detect in traditional tabular data. Knowledge graphs are also used for know-your-customer compliance, where they model the relationships between customers, their business associates, family members, and corporate entities to identify politically exposed persons, sanctions risks, and adverse media connections. In investment research, knowledge graphs enable analysts to understand the complex web of relationships between companies, their competitors, suppliers, customers, and investors. An analyst researching a company can traverse the knowledge graph to identify key business relationships, competitive dynamics, and supply chain dependencies that might affect the company's performance. Knowledge graphs are also used for regulatory compliance, where they model the relationships between regulations, affected business units, required controls, and compliance obligations. This enables compliance teams to quickly identify which regulations apply to which business activities and what controls are required. The integration of knowledge graphs with RAG systems is a particularly powerful combination for financial services. The knowledge graph provides structured, factual grounding for LLM responses, while the RAG system provides the ability to retrieve and synthesize information from unstructured documents. Together, they enable AI systems that can answer complex financial questions with both breadth and precision, grounded in a structured understanding of the financial domain.

Real-World Example

A global bank deploys a knowledge graph for its anti-money laundering investigations. The graph contains 10 million entities including accounts, transactions, individuals, companies, and addresses, with 50 million relationships connecting them. When an investigator identifies a suspicious transaction, the system automatically traverses the knowledge graph to identify related accounts, common signatories, shared addresses, and circular transaction patterns. In one case, the knowledge graph identified a money laundering network involving 50 shell companies, 200 accounts, and 15 individuals that had been active for 3 years. The graph-based analysis, which took seconds, would have required weeks of manual investigation using traditional methods. The bank reports that the knowledge graph has improved suspicious activity detection by 40% and reduced investigation time by 60%. The system also generates automated investigation reports by traversing the graph and presenting the relevant relationships in a visual format that investigators can easily understand and present to regulators.

Why It Matters for Finance

Knowledge graphs are a transformative technology for financial services because they capture the relational structure of the financial world, which is inherently interconnected. Financial crime, business relationships, regulatory requirements, and market dynamics are all fundamentally about connections between entities, and knowledge graphs are the most natural and powerful way to model these connections. The integration of knowledge graphs with LLMs and RAG systems represents the next frontier of financial AI, combining structured factual knowledge with the flexibility of natural language understanding. Financial institutions that invest in knowledge graphs gain a significant advantage in their ability to detect financial crime, understand complex business relationships, and answer sophisticated questions that require traversing multiple domains of knowledge. As regulatory requirements for financial transparency and crime detection continue to increase, knowledge graphs will become an increasingly essential tool for financial institutions.

Explore in Finatune

Collibra Quantexa

Frequently Asked Questions

What is a knowledge graph in finance?

A knowledge graph is a structured representation of financial knowledge that organizes entities like companies, people, and transactions as nodes connected by relationships. It enables sophisticated reasoning, inference, and traversal of connected financial information.

How are knowledge graphs used for financial crime detection?

Knowledge graphs are used for financial crime detection by modeling the relationships between accounts, transactions, individuals, and entities. Investigators can traverse the graph to identify suspicious patterns like circular transactions, layered ownership, and hidden beneficial ownership.

How does a knowledge graph improve financial RAG pipelines?

A knowledge graph improves financial RAG pipelines by providing structured factual grounding for LLM responses. It enables the system to traverse relationships between entities and retrieve information from the graph alongside unstructured document retrieval.

← Previous Term: Knowledge Distillation

Next Term: Large Language Model (LLM) →

View All Fintech Terms →