Today’s Best Practices and Tools: Transforming Financial Data into Knowledge Graphs and LLM Systems

Co-created by the Catalyzer Think Tank divergent thinking and Gemini Deep Research tool.

Introduction: The Transformative Potential of Knowledge Graphs and LLMs in Finance

The financial industry operates on a foundation of vast and intricate data. Financial companies generate and accumulate enormous volumes of data daily from a multitude of sources, encompassing structured information within databases and data warehouses, alongside unstructured data found in documents, reports, and various forms of communication.1 A significant challenge in this landscape is the prevalence of data silos, where critical information remains isolated within specific departments or systems, hindering a comprehensive and unified understanding of the institution’s overall financial picture.2

Traditional relational databases, while robust for managing structured data, often struggle to efficiently represent and query the complex web of interconnected relationships that characterize modern financial data. The many-to-many relationships between customers, accounts, transactions, assets, and market events can be cumbersome to model and navigate using rigid tabular structures.7 Consequently, extracting meaningful insights from this fragmented and complex data often demands substantial manual effort for data aggregation, transformation, and subsequent analysis, leading to operational inefficiencies and an increased potential for errors.2 In today’s dynamic and fast-paced financial environment, the increasing demand for real-time analytics and rapid decision-making further highlights the inherent limitations of these traditional data management approaches.8 Moreover, delays in accessing and interpreting pertinent data can result in missed opportunities and slower reaction times to both market fluctuations and emerging risks.5

To address these evolving challenges, financial institutions are increasingly looking towards next-generation solutions like knowledge graphs and Large Language Models (LLMs). Knowledge graphs offer a paradigm shift by representing financial data, concepts, entities, and their intricate relationships as interconnected networks, moving beyond the constraints of traditional row-and-column structures to capture the inherent meaning within the data.1 This approach organizes information in a manner that closely mirrors how humans naturally understand interconnected concepts.2 Complementing this, Large Language Models (LLMs) are advanced artificial intelligence systems that leverage deep learning models, pre-trained on massive datasets, to understand and generate human-like text, demonstrating remarkable capabilities in various natural language processing tasks.15

Knowledge graphs excel at integrating both structured data from established databases and data warehouses with the wealth of unstructured information found in financial documents and reports, providing a holistic and unified view of an organization’s knowledge assets.1 Simultaneously, LLMs bring powerful capabilities in natural language understanding, sophisticated reasoning, and coherent text generation, enabling them to interact with and derive valuable insights from complex textual data.15 The integration of these two cutting-edge technologies presents a synergistic pathway to overcome the inherent limitations of traditional systems, unlocking the potential for more intelligent data analysis, enhanced insight generation, and ultimately, more informed and effective decision-making.6

The financial industry is under mounting pressure to harness the power of artificial intelligence to gain a crucial competitive edge, significantly improve risk management protocols, foster stronger and more personalized customer relationships, and ensure rigorous adherence to an increasingly complex regulatory landscape. Knowledge graphs and LLMs are rapidly emerging as indispensable tools for financial companies striving to achieve these critical objectives.1

Demystifying Knowledge Graphs for Financial Data

A knowledge graph, in the context of financial data management, can be defined as a knowledge base where factual information is stored in a graph structure rather than traditional tables, with a primary emphasis on illustrating how different data points are interconnected based on their inherent meaning.13 Unlike a simple database, a financial knowledge graph models the financial world as an intricate network of interconnected entities and relationships, providing a structured and highly contextualized lens through which to view complex financial information.2

At its core, a knowledge graph comprises three fundamental components. Nodes, also sometimes referred to as entities or classes, represent the key elements within the financial domain. These can include tangible assets like financial institutions, individual customers, and various types of financial instruments such as stocks and bonds, as well as more abstract concepts like prevailing market trends and evolving regulatory frameworks.1 Edges, also known as relationships or links, define the connections between these nodes, illustrating how different entities and concepts are related to one another. In finance, these relationships might express ownership (“owns”), investment activities (“invests in”), transactional connections (“transacts with”), regulatory oversight (“is regulated by”), or causal influences (“impacts”).1 Finally, properties are attributes or characteristics associated with both the nodes and the edges, providing additional layers of context and detail. For instance, a “Company” node might have properties such as its name, stock ticker symbol, and market capitalization, while a “Transaction” edge could be characterized by properties like the transaction amount, date, and type.7 Consider a practical example: a node representing a major financial institution like “JPMorgan Chase” could be connected by an edge labeled “is regulated by” to another node representing “The Federal Reserve,” with each of these nodes having numerous properties that further describe their characteristics and functions.2

Financial knowledge graphs possess several key characteristics that distinguish them from traditional data management systems. They are fundamentally interconnected networks, representing financial information as a rich tapestry of linked nodes and edges.1 A crucial aspect is their ability to facilitate semantic understanding, going beyond mere keyword matching to capture the underlying meaning of financial terms and relationships, enabling more intelligent and context-aware queries using natural language.1 Furthermore, they enable data enrichment by seamlessly linking related information originating from diverse sources, providing a more holistic and comprehensive view of complex financial ecosystems.1 This includes the capability to integrate both structured data, such as that residing in databases, and unstructured data, like the content of financial reports.1 A core strength lies in their ability to explicitly define and map the intricate relationships between various financial entities, including companies, assets, markets, and transactions.1 Unlike traditional, static databases, knowledge graphs are inherently dynamic, designed to evolve and offer new insights and inferences as new data is incorporated into the system.19 This adaptability allows them to recognize connections between entities without requiring explicit programming for each new piece of information.19 Finally, financial knowledge graphs are adept at handling data originating from diverse sources that often exhibit varying structures, necessitating a collaborative and flexible approach to data structuring that considers schemas, entity identities, and contextual information.19

To further illustrate the nature of financial knowledge graphs, consider a few examples of how financial data can be modeled. Customer relationships can be represented by linking customers to their various accounts, detailing their transactions, outlining their investment portfolios, and tracking their loan history. This model can be further enriched by incorporating external data, such as social media activity or publicly available records, to create a comprehensive view of each customer.3 This interconnected data allows for answering complex questions, such as identifying clients with significant exposure to high-risk markets.3 Another common model involves representing financial instruments like stocks, bonds, and derivatives, along with their dependencies on various counterparties, prevailing market conditions, and relevant regulatory requirements.3 Such a model can help visualize the potential cascading effects of a default in one sector across the broader financial system.3 Knowledge graphs are also effective for modeling company structures, including the relationships between parent companies and their subsidiaries, mapping ownership hierarchies, and identifying key personnel within organizations, linking this information to financial performance and market activities.3 The complex domain of regulatory compliance can be modeled by connecting specific regulatory requirements to the relevant entities, individual transactions, and established internal policies, enabling automated checks for adherence to legal and industry standards.2 Furthermore, knowledge graphs can be used to visualize the intricate relationships between market trends, key economic indicators, and their multifaceted impact on various financial entities like publicly traded companies and diverse asset classes.1 The open-source FinDKG dataset provides a concrete example of a financial knowledge graph schema, featuring 15 predefined relationship types and 12 predefined entity types specifically designed to model the dynamic nature of global economic and market trends over time.33

III. Unlocking Value: Benefits and Use Cases of Knowledge Graphs in Finance

Employing knowledge graphs for financial data management offers a wide array of significant advantages. One of the most prominent benefits is improved data integration, as knowledge graphs provide a unified view of information that originates from disparate sources, effectively breaking down long-standing data silos and enabling a more holistic understanding of financial data across various dimensions.1 This leads to enhanced data accessibility, offering financial professionals self-service access to complex data that resides across numerous systems, allowing them to find relevant information more efficiently without requiring extensive technical expertise or the time-consuming process of sifting through countless spreadsheets.11 Knowledge graphs also increase data interoperability, enabling different systems and even separate organizations to easily share and understand information by creating a common, unified view that facilitates seamless data flow between various departments and systems.2

Furthermore, knowledge graphs significantly improve reasoning capabilities, empowering both human analysts and automated systems to draw more meaningful conclusions from complex financial relationships by identifying previously hidden connections and subtle dependencies within the data.2 They also enhance the context and profundity of other artificial intelligence techniques, particularly machine learning, by providing a structured and semantically rich dataset that can lead to the development of more accurate and insightful predictive models.8 In the realm of regulatory reporting, knowledge graphs facilitate simplified reporting by explicitly capturing the relationships between intricate regulatory definitions and the vast amounts of company data, thereby helping to automate the often cumbersome and highly manual tasks associated with regulatory compliance.2 This leads to better decision-making across various levels of the financial institution by providing a comprehensive and interconnected view of financial transactions, detailed client interactions, prevailing market trends, and potential risks, ultimately enabling more informed and strategic choices.2

The adoption of knowledge graphs also contributes to increased efficiency and automation of numerous financial processes, ranging from the crucial task of data preparation for advanced AI and machine learning applications to the critical functions of fraud detection, compliance checks, and day-to-day customer service operations.2 Ultimately, the enhanced capabilities and efficiencies afforded by knowledge graphs can translate into a significant competitive advantage for financial firms, enabling them to extract greater value from their data, respond more swiftly and effectively to dynamic market changes, and offer innovative financial products and services.1 Moreover, knowledge graphs can diminish the necessity for extensive, labeled datasets that are often required to train AI models and enhance the explainability and transparency of AI-driven insights by providing a clear and structured representation of the underlying knowledge and the reasoning pathways.8

Knowledge graphs are being applied across a spectrum of critical use cases within the financial industry. In risk management, they enable the modeling of complex dependencies between financial instruments, various counterparties, and the overall market conditions, leading to a more nuanced and comprehensive understanding of potential risks.3 Financial institutions can leverage knowledge graphs to visualize how a default or disruption in one specific sector, such as real estate, could potentially cascade through a network of interconnected financial entities, including loans, derivatives, and insurance contracts, thereby aiding in the crucial assessment of systemic risk.3 These sophisticated models also support the execution of stress-testing scenarios and facilitate more informed capital allocation decisions by allowing for the modeling of various risk factors and their potential impacts on the institution’s financial health.3 Furthermore, knowledge graphs are instrumental in identifying and assessing the risks associated with lending and investment opportunities by integrating a wide range of data points, including client behavior patterns, credit scores, and fluctuations in market conditions, providing a more holistic and insightful risk profile for decision-making.26 By mapping the intricate relationships between diverse risk factors, financial institutions can proactively identify potential threats and implement targeted mitigation strategies before these risks escalate, ultimately safeguarding valuable assets and fostering greater trust with their clientele.11 Knowledge graphs can also be utilized to map the complex web of risk relationships that span the entire organization, from granular transaction patterns to overarching regulatory requirements, establishing a more robust and forward-looking risk management framework.2 For instance, a bank can construct a knowledge graph that intricately links a client’s complete financial history with prevailing macroeconomic indicators to more accurately assess the probability of loan defaults.26

In the domain of fraud detection, knowledge graphs provide a powerful lens for identifying suspicious patterns that might otherwise go unnoticed, such as unusual sequences of circular transactions between seemingly unrelated accounts or peculiar connections between various entities.3 They enable the detailed mapping of shared attributes, including phone numbers, addresses, or even devices, across a multitude of accounts, facilitating the discovery of complex fraud rings and the identification of networks of malicious actors.3 By transforming traditional transactional data into a more intuitive social or entity-centric view, knowledge graphs make it easier to pinpoint groups of individuals engaging in frequent, potentially suspicious transactions among themselves, warranting closer scrutiny for fraudulent activities.34 These advanced systems are also adept at detecting specific types of fraud, including the identification of fake identities, instances of credit card fraud, and attempts at money laundering, by meticulously analyzing the relationships between customers, their various transactions, and other pertinent entities.27 Knowledge graphs facilitate the analysis of the intricate and often subtle relationships that exist between numerous accounts, transactions, and diverse entities, uncovering hidden patterns and anomalies that may serve as strong indicators of potential fraudulent schemes.25 As a practical example, a knowledge graph can effectively reveal a sophisticated fraud ring by identifying multiple accounts that share identical phone numbers and addresses, all linked through a series of rapid and unusually structured transactions.3

Regarding customer relationship management, knowledge graphs empower financial institutions to build rich and highly detailed profiles of both individual customers and corporate clients by seamlessly integrating data from a wide range of internal and external sources, leading to enhanced lead enrichment and a much deeper understanding of specific client needs and preferences.2 This deeper understanding can provide a significant competitive edge by enabling AI-driven research that uncovers crucial insights about competitors, informing more strategic business decisions and market positioning.2 By intelligently connecting comprehensive customer data with detailed transaction histories and real-time market insights, knowledge graphs enable banks to offer highly personalized financial advice, ultimately improving customer satisfaction levels and bolstering long-term retention rates.11 These systems facilitate the creation of a comprehensive 360-degree view of each customer by unifying data that is often scattered across disparate channels and operational systems, providing a holistic understanding of all customer interactions, expressed preferences, and observed behaviors.42 When combined with the power of machine learning algorithms, knowledge graphs can analyze individual customer preferences and historical behaviors to create more effective and targeted marketing campaigns and to offer highly precise product and service recommendations, leading to enhanced customer loyalty and significant increases in revenue generation.8 For instance, a financial institution can utilize a knowledge graph to gain a comprehensive understanding of a high-net-worth individual’s complete investment portfolio, their historical spending patterns, and their stated risk tolerance to offer highly personalized wealth management advice and tailored investment opportunities.8

In the critical area of regulatory compliance, knowledge graphs demonstrate exceptional strength in their ability to connect disparate data sources into a cohesive and unified view, enabling financial institutions to effectively track the complex relationships that exist between various entities, specific transactions, and the ever-evolving landscape of regulatory requirements.2 This integrated approach significantly helps banks maintain compliance while simultaneously reducing the substantial manual effort that is typically required to aggregate and meticulously analyze compliance-related data originating from a multitude of different systems.2 Knowledge graphs can be strategically used to automate the often time-consuming and resource-intensive Know Your Customer (KYC) checks by efficiently aggregating relevant data from various sources, including sanctions lists, corporate registries, and detailed transaction histories, thereby flagging any potential violations more efficiently.3 These systems also facilitate the crucial task of tracking fund flows across borders for anti-money laundering (AML) efforts by intricately linking shell companies to their ultimate beneficiaries through complex networks of ownership and transaction records.3 A significant benefit is that knowledge graphs ensure all compliance-related data is readily traceable and easily auditable, greatly simplifying the process of generating comprehensive reports and clearly demonstrating full adherence to all relevant regulatory standards.11 Consider an example: a financial institution can employ a knowledge graph to map all transactions involving a particular client against the most current sanctions lists and all applicable regulatory requirements, automatically flagging any potential instances of non-compliance for immediate further review.3

Finally, in the realm of investment analysis, knowledge graphs drive highly sophisticated processes by mapping the intricate relationships that exist between individual companies, broader market forces, and a wide array of economic indicators, providing a much more comprehensive view of potential investment opportunities and associated risks.2 They empower analysts to identify hidden connections and subtle dependencies that could significantly impact investment decisions, uncovering valuable insights that might remain obscured when using traditional analytical methods.2 Knowledge graphs can be effectively used to analyze prevailing market trends, overall sentiment, and inherent volatility by integrating data from diverse sources such as breaking news articles, active social media conversations, and detailed financial reports, providing a more nuanced and timely understanding of dynamic market forces.8 They also support the development of sophisticated algorithmic trading strategies by enabling the identification of complex patterns and subtle relationships within vast quantities of market data, which can then inform automated trading decisions.39 Furthermore, knowledge graphs can significantly assist in portfolio optimization by allowing investors to gain a deeper understanding of the relationships between different asset classes and to make more informed decisions regarding asset allocation and diversification strategies.1 As an illustration, an investment firm might leverage a knowledge graph to analyze the complex relationships between a company’s historical stock performance, its official regulatory filings, and the sentiment expressed in trending news articles to develop a more profound understanding of potential investment risks and opportunities.2

Large Language Models: A New Era of Financial Intelligence

Large Language Models (LLMs) represent a cutting-edge category of artificial intelligence, characterized by their very large deep learning models that have undergone pre-training on vast amounts of data, often encompassing billions of words sourced from the entirety of the internet, extensive collections of books, a wide range of articles, and numerous other text-rich environments.17 The foundational architecture underpinning most LLMs is the transformer, an intricate neural network design that includes both an encoder and a decoder, each equipped with sophisticated self-attention capabilities. This architecture enables the model to effectively discern and understand the complex relationships that exist between words and phrases within a given sequence of text.17

LLMs possess remarkable abilities in understanding and generating natural language, or text that closely resembles human communication, based on the extensive data they have been trained on through advanced machine learning techniques.15 They are capable of automatically producing text-based content across a diverse array of formats, including creative writing like novels, professional communications such as emails, and even functional code for software applications.15 Their core functionalities span a wide range of natural language processing tasks, including accurate text summarization, fluent translation between languages, comprehensive question answering across various topics, and the creative generation of new content, which can include not only text but also images, musical compositions, and even software code.16

LLMs acquire their profound language understanding by being trained on billions of words from a multitude of sources. During this intensive training process, the models iteratively adjust their internal parameter values until they achieve a high degree of accuracy in predicting the next word in a given sequence of input tokens.16 A key mechanism they employ is the use of word embeddings, where words are transformed into numerical vector representations. This allows the model to understand the contextual meaning of words and to recognize relationships between words and phrases that have similar meanings.17

LLMs hold immense potential for transformative applications within the financial domain. They can be leveraged for sophisticated financial document analysis and summarization, capable of analyzing vast quantities of financial data, including earnings figures, complex financial reports, and detailed financial statements, to extract valuable insights and generate concise summaries for analysts and decision-makers.36 Furthermore, LLMs can power intelligent chatbots and virtual assistants to significantly enhance customer support within financial institutions. These AI-driven conversational agents can answer detailed questions about customer accounts, investment portfolios, or various loan products using natural language, providing round-the-clock support and handling routine inquiries, thus freeing up human agents to focus on more complex customer issues.15 Their ability to understand customer needs and preferences also allows for a more personalized banking experience.37

In the critical area of advanced fraud detection and anomaly identification, LLMs can scrutinize extensive transaction data to identify subtle anomalies and patterns that might indicate potential fraudulent activities. Their capacity to analyze patterns and detect unusual behaviors enables them to proactively identify emerging cybercrime threat campaigns at earlier stages.15 LLMs also hold significant promise for enhanced investment research and analysis. Trained on real-world financial data from sources like Bloomberg and Capital IQ, they can guide investors by processing complex financial data and market trends, conducting sentiment analysis of financial documents, and offering informed portfolio allocation recommendations.36 Their analytical capabilities extend to forecasting market trends and predicting stock price movements with a degree of precision that can surpass traditional methods.36

Finally, LLMs can significantly contribute to the automation of compliance-related tasks within the financial industry. They can assist in navigating the intricate web of regulatory guidelines, complex legal documents, and specific compliance mandates, automating the often-manual process of data collection, improving the speed and accuracy of compliance-related decisions, and even generating predefined templates for various types of financial documents.15

LLMs are poised to usher in a new era of financial intelligence, offering the potential to revolutionize how financial institutions operate, analyze data, interact with customers, manage risk, and ensure regulatory compliance. Their unique ability to understand and generate human language, coupled with their capacity to process and learn from vast datasets, makes them a transformative technology for the financial services sector.

The Synergistic Power: Integrating LLMs with Knowledge Graphs for Financial Insights

The true potential for unlocking advanced financial intelligence lies in the synergistic integration of Large Language Models (LLMs) with knowledge graphs. This powerful combination leverages the distinct strengths of each technology, resulting in enhanced capabilities for financial data analysis that surpass what either can achieve independently. One of the most significant advantages of this integration is the ability to reduce hallucinations, a common issue with standalone LLMs where they may generate factually incorrect information. By grounding LLMs in the verified facts and relationships contained within a knowledge graph, the accuracy and reliability of the AI’s outputs are significantly improved.6

Knowledge graphs provide LLMs with structured, factual information and clearly defined explicit relationships between entities. This acts as a reliable reference source for the LLMs, enhancing their understanding of the financial domain and enabling them to generate more accurate and context-aware responses.21 The integration leverages the inherent semantic richness of embeddings, which capture the meaning of entities and relationships as vectors, and the complex interrelations of concepts within knowledge graphs. This synergy significantly enhances the reasoning abilities of LLMs, allowing them to perform sophisticated tasks like multi-hop reasoning, where they can connect disparate pieces of information to answer complex questions.23 Furthermore, knowledge graphs improve the contextual understanding of LLMs by providing access to external, up-to-date, and domain-specific information, leading to more informed and relevant outputs that are better tailored to the user’s specific needs.24

A crucial aspect of this integration is the ability of contextual LLMs to cite the knowledge graph sources used in their responses. This enhances the transparency and auditability of the AI-generated content, which is particularly important in the highly regulated financial industry where traceability and verification are paramount.24 Compared to the extensive fine-tuning or full training of standalone LLMs for specialized financial tasks, integrating with knowledge graphs can be more cost-effective in terms of both the amount of training data required and the computational resources needed.24 Moreover, the combined system offers greater scalability and flexibility to handle large and complex financial datasets and to process intricate queries that might overwhelm conventional LLMs when dealing with purely textual data.24 By leveraging the interconnected structure of knowledge graphs, LLM-powered systems can uncover non-obvious relationships and provide context-aware suggestions that traditional analytical methods might miss, leading to deeper insights and a broader range of perspectives in financial analysis.21 Finally, the integration is not just about LLMs consuming knowledge graphs; LLMs can also be used to automate the creation of knowledge graphs by intelligently extracting entities and relationships from unstructured financial data, thereby streamlining the often-laborious process of knowledge graph construction.6

One key technique for integrating LLMs and knowledge graphs is Retrieval-Augmented Generation (RAG). In this approach, when an LLM receives a query, it first retrieves relevant information from an external knowledge source, which in this case is a financial knowledge graph, and then uses this retrieved information to augment its response. This provides the LLM with the necessary context and grounding to generate more accurate and informative answers.21 A specialized form of RAG for knowledge graphs is GraphRAG, which goes beyond retrieving simple text snippets and instead focuses on retrieving relevant subgraphs or interconnected paths from the knowledge graph. This allows the LLM to reason over structured, interconnected information about financial entities and their relationships, leading to more contextually rich and logically coherent responses.21 For example, in analyzing financial statements, a GraphRAG system could retrieve the specific subgraph detailing a company’s revenues, expenses, and net income, along with the relationships between these entities, to answer a user’s question about profitability drivers.29

Laying the Foundation: Best Practices for Extracting Financial Data for Knowledge Graphs

The process of transforming a financial company’s data warehouses and databases into knowledge graphs begins with the crucial step of extracting relevant information. Several best practices should be followed to ensure this process is effective and yields high-quality data for the knowledge graph. Initially, it is vital to define clear goals and objectives for the knowledge graph. This will dictate the specific types of information that need to be extracted and the relationships that are important to model.67 Engaging with various stakeholders across the financial institution can help in clearly defining these objectives.67 Following this, it is essential to establish the relevant knowledge domain and the precise scope of data that the knowledge graph will encompass, focusing on the entities, concepts, and their characteristics that are pertinent to the defined goals.67

Financial institutions must then identify all key data sources within their organization that contain information relevant to the defined scope. These sources can include a wide array of systems such as core transaction processing systems, customer relationship management (CRM) platforms, enterprise resource planning (ERP) systems, various market data feeds, comprehensive regulatory filings, and extensive news archives.11 For structured data that resides in relational databases and sophisticated data warehouses, standard SQL queries can be effectively used to retrieve the specific tables, columns, and records that are relevant to the knowledge graph’s intended schema.69 A thorough understanding of the underlying data models of these systems is crucial for constructing accurate and efficient queries. Organizations can also leverage specialized data extraction tools that often provide a variety of pre-built connectors to different types of data sources, which can significantly streamline the process of accessing and retrieving data from disparate and heterogeneous systems.69 When planning the extraction process, it is important to consider the most appropriate strategy: whether to perform a full extraction of all relevant data at the outset or to implement a more iterative incremental extraction process for regularly updating the knowledge graph with any new or modified data.69 Finally, leveraging the established principles of data warehousing can be highly beneficial, as data warehouses typically contain data that has already undergone cleansing, standardization, and integration processes, making it a potentially reliable and high-quality source for populating the knowledge graph.12

A significant aspect of financial data is its diversity in format. Structured data, commonly found in databases and data warehouses, can be directly queried and mapped to the knowledge graph’s nodes and edges based on a well-defined schema.71 Semi-structured data, such as JSON and XML files often used for data interchange, can be parsed and transformed into the graph model by extracting relevant information based on the inherent structure of these files.72 Unstructured data, which includes a vast amount of information in the form of text documents like financial reports, news articles, and regulatory filings, necessitates the application of Natural Language Processing (NLP) techniques. These techniques are used to identify and extract key entities, relationships, and underlying concepts that can then be accurately represented within the knowledge graph.8 This process often involves sophisticated methods such as named entity recognition and relation extraction. For financial information contained in scanned documents and PDF files, Optical Character Recognition (OCR) technology can be employed to convert the images of text into machine-readable formats, which can then be effectively processed using NLP techniques.69 Similarly, specialized document parsing techniques can be utilized to extract specific, targeted information from financial documents based on their inherent layout and structural elements, such as accurately extracting data from complex tables embedded within a report.69 Increasingly, Large Language Models (LLMs) are being leveraged for highly sophisticated entity and relationship extraction from unstructured financial text. Their advanced understanding of the context and nuances of financial language allows for more accurate and comprehensive information extraction.6

Once the financial data has been extracted, ensuring its quality is paramount. Data cleaning is a critical process that involves identifying and rectifying any errors, inconsistencies, and redundancies present in the data. This may include tasks such as removing duplicate records, effectively handling missing values, and correcting any data entry errors that may have occurred.41 Normalization is essential to ensure that the financial data is consistent and readily comparable across different sources. This can involve standardizing various data formats, including date formats, currency symbols, and units of measure, as well as establishing uniform naming conventions across all the integrated data sources.71 Entity resolution is a vital technique used to identify and accurately link different records or mentions that refer to the same real-world entity. For example, the same company might be referred to by slightly different names or abbreviations in various datasets. Techniques such as advanced record linkage, sophisticated fuzzy matching algorithms, and machine learning-based methods can be effectively employed for this crucial task.71 Establishing and adhering to consistent naming conventions for all entities and relationships within the knowledge graph is also crucial to avoid any ambiguity and to ensure the overall integrity of the data.75 Implementing robust data validation processes at multiple stages throughout the data extraction and loading pipeline helps to ensure the ongoing accuracy of the data and its adherence to the defined schema and underlying ontology of the knowledge graph.72 Finally, Large Language Models (LLMs) can play an increasingly important role in entity disambiguation, helping to accurately determine the specific entity being referenced in different textual contexts, and in merging duplicate entities based on their semantic similarity and the surrounding contextual information.73

VII. Building and Maintaining Robust Financial Knowledge Graphs

Constructing a robust financial knowledge graph is a multifaceted process that requires careful planning and execution. The initial step involves clearly defining the goals and objectives of the knowledge graph, ensuring a shared understanding of its purpose and the specific business questions it aims to address.67 Following this, it is crucial to identify the relevant knowledge domain and precisely define the scope of information that the graph will encompass, focusing on the key financial entities and their pertinent relationships.67 The next phase is data collection and preprocessing, which includes gathering relevant data from identified sources, performing necessary cleaning and transformation steps, and organizing it into a format suitable for knowledge graph construction.67

A critical aspect of building a knowledge graph is semantic data modeling, where the structure of the graph is designed. This involves identifying the core entities (nodes), defining their characteristics (properties), and specifying the nature of the connections (edges or relationships) between them.75 This process often includes the creation or reuse of a formal ontology.72 Selecting an appropriate graph database management system (DBMS) is essential for efficiently storing and querying the knowledge graph. Options include property graph databases like Neo4j and Amazon Neptune, and RDF triple stores such as Ontotext GraphDB.72 The choice of DBMS should be guided by factors such as the intended data model, specific query requirements, and the anticipated scalability needs of the application. Data ingestion is the process of loading the prepared data into the chosen graph database, which can be achieved using ETL (Extract, Transform, Load) tools or through data virtualization techniques that allow for accessing data directly from its source.72 While data modeling defines the conceptual structure, creating a formal ontology provides a semantic blueprint for the knowledge graph, explicitly defining the classes of entities, the types of relationships, and the rules governing them within the financial domain. Leveraging existing, well-established ontologies like FIBO (Financial Industry Business Ontology) can provide a significant advantage by offering a standardized vocabulary and semantic framework.76 The construction process should ideally follow a phased implementation approach, beginning with a small, representative subset of the data to validate the initial model and then progressively expanding the graph as needed.76 Finally, thorough testing and validation are crucial to ensure the accuracy, consistency, and overall quality of the constructed knowledge graph. This involves running a variety of queries and carefully validating the resulting information against expected outcomes and the expert knowledge of financial domain specialists.72

For effective data modeling and ontology creation, it is best to begin by clearly identifying the core financial entities and the essential relationships that exist between them, always keeping in mind the specific questions that the knowledge graph is intended to answer.72 Creating a well-defined ontology is paramount for providing a formal and unambiguous representation of the knowledge within the graph. This involves meticulously defining the different types of entities (known as classes), their associated properties (attributes), and the various types of relationships that can connect these entities.5 It is often beneficial to leverage existing specifications and established standards such as RDF (Resource Description Framework) and OWL (Web Ontology Language) when creating ontologies, as these provide a common, interoperable framework for knowledge representation.28 A particularly valuable resource in the financial domain is the Financial Industry Business Ontology (FIBO), which offers a standardized vocabulary and a comprehensive semantic framework specifically designed for the financial industry, greatly facilitating data harmonization and ensuring unambiguous meaning across different systems and organizations.5 When developing an ontology, it is important to define a clear and focused domain and scope to avoid unnecessary complexity and to ensure that the ontology remains relevant to its intended purpose.91 A highly effective approach involves combining the expertise of financial domain specialists with the advanced capabilities of LLMs for both the initial creation and the ongoing refinement of the ontology. LLMs can assist in automatically identifying key concepts and relationships from the data, while human experts can provide crucial validation and add essential contextual knowledge that the AI might miss.91 Throughout the process, it is critical to focus on maintaining semantic consistency across the entire ontology and to clearly define the nature of the relationships between entities to ensure that the meaning of the financial data is accurately and unambiguously represented.91 Finally, it is often advisable to start with a relatively small and focused ontology, addressing the most critical concepts and relationships first, and then to grow the ontology iteratively based on the evolving needs of the users and the increasing complexity of the data.80

The selection of a suitable graph database technology is a critical decision for any financial knowledge graph implementation. Graph databases, in general, are specifically designed and optimized for efficiently storing and querying data based on the intricate relationships between data points, making them ideally suited for the interconnected nature of knowledge graphs.7 There are two primary types of graph databases to consider: property graph databases and RDF triple stores. Property graph databases, such as Neo4j, Amazon Neptune, and TigerGraph, store data as nodes and edges, where both nodes and edges can have associated properties. Neo4j is a widely adopted choice, known for its native graph storage architecture and its intuitive Cypher query language.3 Amazon Neptune is a fully managed graph database service offered by Amazon Web Services (AWS) that supports both property graphs and RDF.3 TigerGraph is recognized for its high-performance capabilities in analyzing connected data in real time.42 RDF triple stores, including Ontotext GraphDB and Amazon Neptune, store information as subject-predicate-object triples, which aligns directly with the fundamental structure of knowledge graphs. Ontotext GraphDB is particularly noted for its high performance and scalability when managing large RDF datasets and supports the standardized SPARQL query language.3 When selecting a graph database for financial applications, it is essential to carefully consider factors such as the anticipated scalability requirements to handle potentially massive datasets, the necessary performance levels for real-time queries and complex analytics, the strength of security features to protect sensitive financial information from unauthorized access, and the overall compatibility with the organization’s existing IT infrastructure.2 Other graph database technologies like ArangoDB, Dgraph, JanusGraph, and Memgraph may also be suitable depending on the specific needs and priorities of the financial institution.77

Maintaining a financial knowledge graph is an ongoing endeavor that requires consistent effort and strategic planning. It is crucial to establish a process for regularly updating the data within the knowledge graph to ensure that the information remains accurate, relevant, and reflects the most current state of the financial landscape.11 Implementing automated data pipelines can significantly streamline this process by automatically ingesting new data from various identified sources and updating the knowledge graph in real time as new information becomes available, thereby reducing the need for manual intervention and ensuring data freshness.110 To guarantee the reliability and trustworthiness of the knowledge contained within the graph, it is essential to utilize robust quality control measures. This can involve employing machine learning algorithms to automatically detect and correct potential errors and inconsistencies, as well as establishing manual review processes conducted by financial domain experts to validate the accuracy of the information.41 To maximize its utility and value, the knowledge graph should be integrated with other relevant tools and systems used across the financial institution, such as advanced data analytics platforms, internal knowledge management software, and other pertinent data sources.110 It is also important to continuously monitor the performance of the knowledge graph, tracking metrics such as query execution times and identifying any areas for potential optimization. This ensures that the graph remains efficient and responsive as the volume of data grows and the complexity of queries increases.110 As the financial landscape is constantly evolving, it is vital to develop strategies for adapting the knowledge graph to market changes and any evolving regulatory requirements, ensuring that the information it contains remains relevant and compliant with the latest industry standards and legal frameworks.2 Finally, proactively addressing data governance challenges by establishing clear and well-defined policies and procedures for data ownership, access control, and overall data quality within the knowledge graph environment is crucial for its long-term success and sustainability.2

VIII. Intelligent Interaction: Connecting and Querying Knowledge Graphs with LLM Systems

Effectively leveraging the power of both knowledge graphs and Large Language Models (LLMs) in finance requires robust techniques for connecting and enabling seamless interaction between these two advanced technologies. One of the primary methods for achieving this is through Retrieval-Augmented Generation (RAG). In this framework, when an LLM receives a user query, it first retrieves relevant information from an external knowledge source, such as a financial knowledge graph, and then uses this retrieved information to augment the generation of its response. This process provides the LLM with the necessary context and grounding to produce more accurate and informative answers.21 A specialized adaptation of RAG specifically designed for knowledge graphs is GraphRAG. Instead of retrieving unstructured text documents or isolated data chunks, GraphRAG focuses on retrieving relevant subgraphs or interconnected paths directly from the knowledge graph. This allows the LLM to reason over structured, semantically rich information about financial entities and their intricate relationships, leading to more contextually accurate and logically coherent responses.21

Conversely, the integration is not unidirectional. LLMs can also be effectively used to automate the creation of knowledge graphs from the vast amounts of unstructured financial data that exist within an organization. By employing sophisticated techniques like named entity recognition, relationship extraction, and contextual inference, LLMs can automatically transform free-form text into structured knowledge graph representations, significantly streamlining the graph construction process.6 Specialized tools known as LLM graph transformers further facilitate this process by enabling users to define prompts that guide the LLM in extracting specific entities and relationships from text, which are then stored as nodes and edges in a graph database.6 Often, the most effective solutions involve creating hybrid systems that strategically combine the strengths of symbolic AI, represented by the structured knowledge and reasoning capabilities of knowledge graphs, with the advanced natural language processing and generation abilities of generative AI, embodied by LLMs.24 In this paradigm, knowledge graphs serve as a reliable and factual foundation for LLMs, enriching them with verified data and leading to significantly improved reasoning capabilities and a deeper contextual understanding of complex financial topics.21

To enable intuitive interaction with financial knowledge graphs, several methods exist for querying the graph using natural language, powered by LLMs. Natural Language Querying (NLQ) allows users to ask questions about the financial data stored in the knowledge graph using their own words, eliminating the need to learn and write complex formal query languages like SPARQL or Cypher.22 LLMs play a pivotal role in NLQ by translating these natural language queries into the specific query language that the underlying graph database understands, such as SPARQL for RDF-based graphs or Cypher for Neo4j.22 One prominent technique for this is Text2Cypher, which focuses specifically on converting natural language questions into Cypher queries that can be directly executed against a Neo4j graph database, allowing users to access and explore the graph’s information without needing any prior knowledge of Cypher syntax.22 For more intricate financial inquiries that involve multiple interconnected pieces of information or require complex reasoning steps, LLM agents can be employed using a chain-of-thought approach. This involves breaking down the initial question into a series of smaller, more manageable sub-questions that the agent then uses to navigate the knowledge graph and progressively build towards a comprehensive answer.61 Furthermore, the integration of LLMs with knowledge graphs can facilitate the creation of interactive graph dashboards where users can pose questions in natural language, and the LLM will translate these into the appropriate queries to dynamically visualize the relevant financial data stored within the graph.22

Finally, semantic embeddings offer another powerful approach for enhancing information retrieval and enabling sophisticated reasoning over financial data within knowledge graphs. Knowledge graph embeddings are essentially vector representations of the entities and the relationships within the graph, capturing their underlying semantic meaning and structural context in a continuous, multi-dimensional vector space.122 These embeddings can be utilized to predict missing links within the knowledge graph, helping to automatically complete and enrich the financial knowledge base with inferred relationships.122 Various graph embedding techniques, such as DeepWalk and Node2Vec, can be applied to financial knowledge graphs to learn vector representations of nodes based on their surrounding context and their position within the network, effectively capturing both local and broader structural information.125 Graph Neural Networks (GNNs) leverage these graph embeddings to perform a variety of machine learning tasks on the knowledge graph, including the critical application of fraud detection by identifying subtle anomalies and suspicious patterns within financial transaction networks.27 Semantic embeddings also enable semantic search capabilities within the knowledge graph, allowing users to retrieve information based on the meaning and intent of their queries rather than just relying on exact keyword matches, leading to more relevant and insightful results for financial analysis.122 Moreover, the integration of knowledge graph embeddings with the dense vector representations used by LLMs can significantly enhance the reasoning abilities of the LLMs when processing financial data, enabling them to perform more complex inferences and answer nuanced questions with greater accuracy.23

Navigating Critical Considerations: Data Security, Privacy, and Compliance in Finance

The implementation of knowledge graphs and Large Language Models (LLMs) within the financial industry, while offering significant advantages, necessitates a thorough and careful consideration of several critical factors, most notably data security, privacy, and regulatory compliance. Given the highly sensitive nature of financial information, addressing these aspects with robust strategies and best practices is paramount.

Financial institutions must proactively address the significant data security challenges that arise when implementing knowledge graphs and LLMs. This includes establishing enterprise-grade security measures such as stringent access controls based on roles and responsibilities, and comprehensive data protection mechanisms designed to prevent unauthorized access, use, or disclosure of sensitive financial data.2 A key strategy to consider is federating data sources rather than undertaking massive and potentially risky data migrations. This approach allows sensitive data to remain within its existing, secured environments while still enabling the AI systems to access and process it as a unified and logically connected source.4 It is also crucial to implement robust safeguards to protect LLMs from various vulnerabilities, including the risk of data exfiltration, where confidential information could be inadvertently or maliciously leaked; the potential for unauthorized code execution through carefully crafted prompts; the possibility of fraudulent activities being triggered by manipulated model outputs; and the significant threat of reputational damage that could result from inappropriate or inaccurate content generated by the AI.134 To mitigate these risks, organizations should implement prompt security controls to prevent prompt injection attacks, utilize sandboxing and segmentation techniques to isolate LLM processes and limit their access to critical systems, integrate behavioral analysis to detect any anomalous or suspicious activities, and maintain rigorous audit trails of all interactions with the LLM systems.136 Furthermore, encrypting data both while it is being stored (at rest) and during transmission (in transit) is a fundamental best practice to ensure that sensitive financial information remains protected from unauthorized interception or access.134 LLMs should be kept up to date with the latest security patches to address any newly discovered vulnerabilities and ensure the models remain resilient against evolving cyber threats.134 Finally, employing multi-factor authentication (MFA) and role-based access control (RBAC) are essential for limiting access to LLM systems and the underlying financial data to only those personnel who have a legitimate business need.134

Addressing privacy considerations is equally critical when dealing with financial data in knowledge graphs and LLM systems. Financial institutions must ensure controlled access to any information that relates to individual customers or entities to comply with increasingly stringent privacy regulations and to maintain the trust of their clientele.137 Implementing granular access controls within the knowledge graph allows for the definition of precise permissions regarding who can access specific data elements and the relationships between them, thereby respecting established privacy boundaries.138 To further protect sensitive information, organizations should employ data anonymization and redaction techniques to remove or mask any personally identifiable information (PII) from the data that is used to train LLMs and to populate knowledge graphs. This significantly minimizes the risk of inadvertent privacy breaches.138 Exploring privacy-enhancing technologies such as blockchain could also be beneficial, as blockchain can provide data immutability and a transparent audit trail while allowing participants in a decentralized knowledge graph network to maintain control over who has access to their specific data.141 When working with LLMs, applying differential privacy techniques to the data used for training and for generating insights can be an effective way to add a controlled amount of noise to the data, making it extremely difficult to identify individual data points while still allowing for valuable analytical insights to be derived.134 Finally, it is crucial to maintain transparency with users regarding how their data is being collected, processed, and utilized within these advanced systems, and to provide them with control over their data where applicable, ensuring full compliance with all relevant data privacy laws and fostering a strong sense of trust.139

Ensuring compliance with relevant financial regulations and industry standards is a non-negotiable requirement for any financial institution implementing knowledge graphs and LLMs. Knowledge graphs can significantly simplify the often-complex process of ensuring compliance by providing a clear and easily traceable view of data lineage and usage, making it much easier to audit and demonstrate full adherence to a wide range of regulatory requirements.11 They can also facilitate the automation of compliance monitoring by continuously tracking any changes in regulatory requirements and assessing potential risk factors based on the interconnected data residing within the graph.14 Financial institutions must ensure strict compliance with key regulations such as GDPR (General Data Protection Regulation), CCPA (California Consumer Privacy Act), and various industry-specific guidelines, including KYC (Know Your Customer) and AML (Anti-Money Laundering) requirements, when designing and deploying knowledge graphs and LLMs.3 Knowledge graphs can aid in mapping specific transactions and entities to the relevant laws and regulatory guidelines, providing a structured framework for understanding and adhering to the complex regulatory landscape.35 Leveraging established industry-standard ontologies such as FIBO (Financial Industry Business Ontology) can be particularly beneficial, as FIBO provides a common and consistent vocabulary that aligns with many regulatory definitions and classifications, thereby simplifying the often-challenging process of compliance reporting and ensuring a shared understanding of critical financial terms.5 Finally, it is crucial to establish clear and comprehensive ethical AI policies that emphasize principles of fairness, transparency, and accountability in the development and deployment of LLMs within the financial sector. This proactive approach helps to meet regulatory expectations and to maintain the trust and confidence of both customers and the broader public.143

Conclusion: Charting the Future of Intelligent Financial Data Management

The integration of knowledge graphs and Large Language Models (LLMs) represents a transformative shift in the way financial institutions manage, analyze, and derive value from their vast data assets. By overcoming the limitations of traditional data management systems, this powerful synergy unlocks unprecedented opportunities for enhanced efficiency, deeper insights, and more informed decision-making across the financial services landscape. Knowledge graphs provide a structured and semantically rich representation of complex financial data, while LLMs offer intuitive natural language interaction and advanced reasoning capabilities.

The adoption of these technologies brings numerous benefits, including improved data integration, enhanced data accessibility, increased data interoperability, and superior reasoning capabilities. Financial companies can leverage knowledge graphs and LLMs to revolutionize critical functions such as risk management, fraud detection, customer relationship management, regulatory compliance, and investment analysis. Best practices for this integration involve a well-defined strategy for data extraction, robust methods for building and maintaining knowledge graphs, and careful consideration of data security, privacy, and compliance from the outset.

Looking ahead, the future of intelligent financial data management will likely see even closer integration between knowledge graphs and LLMs. Advancements in graph embedding techniques will further enhance the ability of AI systems to reason over complex financial relationships. The development of more sophisticated natural language querying interfaces will democratize access to financial insights for a wider range of users. As these technologies continue to evolve, financial institutions that strategically embrace the synergistic power of knowledge graphs and LLMs will be well-positioned to navigate the complexities of the modern financial world, gain a sustainable competitive advantage, and deliver greater value to their stakeholders.

Key Tables:

Key Characteristics of Financial Knowledge Graphs

Characteristic	Description	Application in Finance
Interconnected Networks	Financial information represented as a network of linked entities and concepts.	Modeling relationships between customers, accounts, transactions, and market events.
Semantic Understanding	Captures the meaning of financial data, enabling context-aware queries using natural language.	Asking questions about financial performance or regulatory requirements in plain English.
Data Enrichment	Integrates related information from diverse sources for a holistic view of financial ecosystems.	Combining market data, news sentiment, and company financials for investment analysis.
Relationship Mapping	Explicitly defines and analyzes complex relationships between financial entities.	Identifying connections between shell companies and beneficiaries for anti-money laundering (AML).
Dynamic Nature	Evolves and offers new insights as new financial data is incorporated.	Automatically recognizing emerging risk patterns based on the latest market activities.
Handling Diverse Data	Can structure and integrate data from various sources with differing formats and schemas.	Combining structured transaction data with unstructured news articles about the same financial event.

Key Use Cases of Knowledge Graphs in Finance

Use Case	Description	Example Application
Risk Management	Models dependencies between financial instruments, counterparties, and market conditions to visualize and assess potential risks.	A bank visualizing how a default in the real estate sector could impact its portfolio of loans and derivatives.
Fraud Detection	Identifies suspicious patterns and relationships in transaction data to uncover fraudulent activities and networks.	Mapping shared addresses, phone numbers, and transaction patterns across multiple accounts to detect potential fraud rings.
Customer Relationship Management	Creates a unified view of customer data, including interactions, preferences, and financial holdings, to enable personalized services.	A wealth management firm using a knowledge graph to understand a client’s investment history, risk tolerance, and financial goals to provide tailored investment advice and product recommendations.
Regulatory Compliance	Connects regulatory requirements with relevant entities, transactions, and internal policies to automate compliance checks and reporting.	A financial institution using a knowledge graph to automatically flag transactions involving entities on sanctions lists and to track compliance with Know Your Customer (KYC) regulations.
Investment Analysis	Maps relationships between companies, markets, and economic indicators to identify hidden connections and inform investment decisions.	An investment analyst using a knowledge graph to analyze the relationships between a company’s financial performance, its supply chain, and global market trends to gain deeper insights for investment strategies.

Potential Applications of LLMs in Finance

Application	Description	Benefits
Financial Document Analysis and Summarization	Analyzing complex financial reports, earnings calls, and regulatory filings to extract key information and generate concise summaries.	Saves time for analysts, improves efficiency in information processing, and enables quicker access to critical insights.
Intelligent Chatbots and Virtual Assistants	Answering customer inquiries about accounts, products, and services in natural language, providing 24/7 support.	Enhances customer service, improves response times, reduces workload for human agents, and offers personalized support.
Advanced Fraud Detection and Anomaly Identification	Analyzing transaction patterns and identifying unusual behaviors to detect and prevent fraudulent activities in real-time.	Proactively identifies potential fraud, reduces financial losses, enhances security, and improves the ability to detect sophisticated fraud schemes.
Enhanced Investment Research and Analysis	Processing market data, news sentiment, and financial reports to guide investment decisions and forecast market trends.	Provides investors with timely and comprehensive insights, supports the development of more effective trading strategies, and enables better prediction of market movements.
Automation of Compliance-Related Tasks	Assisting with navigating regulatory guidelines, automating data collection, and generating reports for compliance purposes.	Streamlines compliance processes, reduces manual effort, improves the accuracy of compliance decisions, and helps ensure adherence to complex regulatory frameworks.

Suitable Graph Database Technologies for Financial Applications

Technology	Data Model	Scalability	Performance	Query Language	Suitability for Finance
Neo4j	Property Graph	Highly scalable for large datasets	Optimized for relationship traversal	Cypher	Widely used in finance for fraud detection, risk analysis, customer 360, and knowledge management due to its strong focus on relationships.
Ontotext GraphDB	RDF Triple Store	Excellent scalability for massive RDF data	High performance for semantic queries	SPARQL	Well-suited for applications requiring semantic reasoning, complex metadata management, and integration with semantic web standards, relevant for regulatory compliance and data governance in finance.
Amazon Neptune	Property & RDF	Highly scalable and available on AWS	Fast querying of graph data	Cypher, SPARQL	A fully managed service, suitable for various financial use cases including fraud detection, identity resolution, and building knowledge graphs, with the benefits of AWS infrastructure.
TigerGraph	Property Graph	Designed for massive parallel processing	Fastest for deep-link analytics in real-time	GSQL	Ideal for applications requiring real-time analysis of highly interconnected financial data, such as complex fraud pattern detection and risk assessment across vast networks.

Security Considerations for Financial Knowledge Graphs and LLMs

Challenge	Best Practice
Data Leaks	Implement robust access controls, encrypt data at rest and in transit, anonymize sensitive data, and use federated data management where possible.
Model Manipulation	Regularly update models with security patches, perform integrity checks, and run models in secure execution environments.
Unauthorized Access	Implement multi-factor authentication (MFA) and role-based access control (RBAC) to limit access to authorized personnel only.
Prompt Injection (for LLMs)	Implement prompt security controls, including content filtering on both input and output, and use defensive prompt engineering techniques.
Data Poisoning (for LLMs)	Use clean and unbiased datasets for training, encrypt and validate all data, and continuously monitor for data manipulation attempts.
Compliance Violations	Adhere to data protection laws (GDPR, CCPA, etc.), implement data localization where required, maintain audit trails, and establish ethical AI policies.
Sensitive Information Disclosure (by LLMs)	Use data sanitization techniques, implement strict user access policies, and continuously monitor LLM outputs for potential leaks.
Supply Chain Vulnerabilities	Regularly review and update all dependencies, including models and code libraries, to address any potential security weaknesses.

Works cited

What is Knowledge Graphs in Finance? – Kosh.ai, accessed April 3, 2025, https://www.kosh.ai/glossary-words/knowledge-graphs-in-finance
Knowledge Graphs in Finance: Revolutionizing Financial Data Analysis – SmythOS, accessed April 3, 2025, https://smythos.com/ai-industry-solutions/finance/knowledge-graphs-in-finance/
How can knowledge graphs be applied in the financial industry? – Milvus, accessed April 3, 2025, https://milvus.io/ai-quick-reference/how-can-knowledge-graphs-be-applied-in-the-financial-industry
Knowledge Graphs are Critical to Trustworthy Enterprise AI – Stardog, accessed April 3, 2025, https://www.stardog.com/blog/knowledge-graphs-are-critical-to-trustworthy-enterprise-ai/
The Power of Knowledge Graphs within the Financial Industry – Ontotext, accessed April 3, 2025, https://www.ontotext.com/blog/the-power-of-knowledge-graphs-within-the-financial-industry/?generate_pdf=43875
Knowledge Graphs with LLMs: Optimizing Decision-Making – Addepto, accessed April 3, 2025, https://addepto.com/blog/leveraging-knowledge-graphs-with-llms-a-business-guide-to-enhanced-decision-making/
What is a Knowledge Graph? – Esri, accessed April 3, 2025, https://www.esri.com/arcgis-blog/products/arcgis-enterprise/data-management/what-is-a-knowledge-graph/
Knowledge Graphs: Transforming Personalized Finance – Fractal Analytics Inc., accessed April 3, 2025, https://fractal.ai/knowledge-graphs-transforming-personalized-finance/
Leveraging Graph Databases for Fraud Detection in Financial Systems, accessed April 3, 2025, https://cacm.acm.org/blogcacm/leveraging-graph-databases-for-fraud-detection-in-financial-systems/
4. The Relational Data Warehouse – Deciphering Data Architectures [Book] – O’Reilly, accessed April 3, 2025, https://www.oreilly.com/library/view/deciphering-data-architectures/9781098150754/ch04.html
What role does Knowledge Graph play in finance? – BytePlus, accessed April 3, 2025, https://www.byteplus.com/en/topic/488917
The Benefits of Data Warehousing in Finance, accessed April 3, 2025, https://www.dataideology.com/the-benefits-of-data-warehousing-in-finance/
nicola-76063.medium.com, accessed April 3, 2025, https://nicola-76063.medium.com/knowledge-graphs-and-data-governance-4798f85e8b8c#:~:text=Generally%2C%20a%20knowledge%20graph%20is,different%20data%20points%20are%20connected.
Knowledge Graphs and Data Governance | by Nicola Askham – Medium, accessed April 3, 2025, https://nicola-76063.medium.com/knowledge-graphs-and-data-governance-4798f85e8b8c
What are large language models (LLMs)? – Microsoft Azure, accessed April 3, 2025, https://azure.microsoft.com/en-us/resources/cloud-computing-dictionary/what-are-large-language-models-llms
What is a large language model (LLM)? – SAP, accessed April 3, 2025, https://www.sap.com/resources/what-is-large-language-model
What is LLM? – Large Language Models Explained – AWS, accessed April 3, 2025, https://aws.amazon.com/what-is/large-language-model/
What Are Large Language Models (LLMs)? – IBM, accessed April 3, 2025, https://www.ibm.com/think/topics/large-language-models
Knowledge graph: Your key to get insights from big data – Finextra Research, accessed April 3, 2025, https://www.finextra.com/blogposting/26329/knowledge-graph-your-key-to-get-insights-from-big-data
What is an LLM (large language model)? – Cloudflare, accessed April 3, 2025, https://www.cloudflare.com/learning/ai/what-is-large-language-model/
Applications of Knowledge Graphs in LLMs: 3 Important Cases – Data Science Dojo, accessed April 3, 2025, https://datasciencedojo.com/blog/knowledge-graphs/
Unifying LLMs & Knowledge Graphs for GenAI: Use Cases & Best Practices – Neo4j, accessed April 3, 2025, https://neo4j.com/blog/genai/unifying-llm-knowledge-graph/
The Synergy Between Knowledge Graphs and Large Language Models – Datanami, accessed April 3, 2025, https://www.bigdatawire.com/2024/05/01/the-synergy-between-knowledge-graphs-and-large-language-models/
Building Accountable LLMs with Knowledge Graphs – Valkyrie AI, accessed April 3, 2025, https://valkyrie.ai/post/building-accountable-llms-with-knowledge-graphs/
Knowledge Graph Use Cases Driving Innovation in 2025 – PingCAP, accessed April 3, 2025, https://www.pingcap.com/article/knowledge-graph-use-cases-2025/
How can knowledge graphs be applied in the financial industry? – Zilliz Vector Database, accessed April 3, 2025, https://zilliz.com/ai-faq/how-can-knowledge-graphs-be-applied-in-the-financial-industry
Fraud detection using knowledge graph: How to detect and visualize fraudulent activities, accessed April 3, 2025, https://www.nebula-graph.io/posts/fraud-detection-using-knowledge-and-graph-database
What Is a Knowledge Graph? | Ontotext Fundamentals, accessed April 3, 2025, https://www.ontotext.com/knowledgehub/fundamentals/what-is-a-knowledge-graph/
Transforming Financial Statements into Knowledge Graphs Using Neo4j LLM Knowledge Graph Builder | by Kshitij Kutumbe, accessed April 3, 2025, https://kshitijkutumbe.medium.com/transforming-financial-statements-into-knowledge-graphs-using-neo4j-llm-knowledge-graph-builder-418a1379c6a8
Top 9 Knowledge Graphs Use Cases, accessed April 3, 2025, https://cdn.prod.website-files.com/625447c67b621ab49bb7e3e5/67a3c0688035b75e2f4ca37a_pdf-knowledge%20graph%20use%20cases.pdf
milvus.io, accessed April 3, 2025, https://milvus.io/ai-quick-reference/how-can-knowledge-graphs-be-applied-in-the-financial-industry#:~:text=Knowledge%20graphs%20can%20enhance%20financial,tasks%20requiring%20context%2Daware%20analysis.
Knowledge Graph Use Cases In Finance – Restack, accessed April 3, 2025, https://www.restack.io/p/knowledge-graph-answer-finance-use-cases-cat-ai
FinDKG: Financial Dynamic Knowledge Graph – Victor Xiaohui Li, accessed April 3, 2025, https://xiaohui-victor-li.github.io/FinDKG/
Why Knowledge Graph for Financial Services? Real Use Cases – Datanami, accessed April 3, 2025, https://www.bigdatawire.com/2022/03/28/why-knowledge-graph-for-financial-services-real-use-cases/
Ontologies and Knowledge Graphs: Unraveling Complex Data in Financial Services, accessed April 3, 2025, https://deepsee.ai/knowledge-center/ontologies-and-knowledge-graphs-unraveling-complex-data-in-financial-services/
Large Language Models in Finance: Benefits and Use Cases – Aisera, accessed April 3, 2025, https://aisera.com/blog/large-language-models-in-financial-services-banking/
The transformative potential of Gen-AI and LLMs in Banking – KPMG International, accessed April 3, 2025, https://assets.kpmg.com/content/dam/kpmgsites/uk/pdf/2023/11/transformative-potential-of-gen-ai-and-llm-in-banking.pdf
Large Language Models in Banking: What Are They Good For? – Dataiku blog, accessed April 3, 2025, https://blog.dataiku.com/generative-ai-in-banking
milvus.io, accessed April 3, 2025, https://milvus.io/ai-quick-reference/how-can-knowledge-graphs-be-applied-in-the-financial-industry#:~:text=One%20key%20use%20case%20is,connections%20between%20seemingly%20unrelated%20entities.
ArcGIS Knowledge for Fraud Detection, accessed April 3, 2025, https://storymaps.arcgis.com/stories/e636b6d1e9aa4a1e95a5aa721d96b04f
Step-by-Step Guide to Building a Knowledge Graph in 2025 – PageOn.ai, accessed April 3, 2025, https://www.pageon.ai/blog/knowledge-graph
Transform Financial Services With TigerGraph Databases and AI, accessed April 3, 2025, https://www.tigergraph.com/solutions/financial-services/
How Knowledge Graph Transforms Banking: BytePlus ModelArk Insights, accessed April 3, 2025, https://www.byteplus.com/en/topic/489066
Large Language Models Applications in Financial Services – KMS Solutions, accessed April 3, 2025, https://kms-solutions.asia/blogs/large-language-models-in-financial-services
Knowledge Graph Explained | The Future of Smart Data – GraphAware, accessed April 3, 2025, https://graphaware.com/glossary/knowledge-graph/
Past & future use cases of Knowledge Graphs and AI integration – Datavid, accessed April 3, 2025, https://datavid.com/blog/knowledge-graphs-and-ai-integration
Knowledge Graphs in Financial Services: How to Handle Disparate Data – YouTube, accessed April 3, 2025, https://www.youtube.com/watch?v=DDoN37DUGo4
A use of knowledge graph – Fordham University Faculty, accessed April 3, 2025, https://faculty.fordham.edu/rchen/JFS.pdf
aws.amazon.com, accessed April 3, 2025, https://aws.amazon.com/what-is/large-language-model/#:~:text=help%20with%20LLMs%3F-,What%20are%20Large%20Language%20Models%3F,decoder%20with%20self%2Dattention%20capabilities.
Large Language Models: What You Need to Know in 2025 | HatchWorks AI, accessed April 3, 2025, https://hatchworks.com/blog/gen-ai/large-language-models-guide/
5 key features and benefits of large language models | The Microsoft Cloud Blog, accessed April 3, 2025, https://www.microsoft.com/en-us/microsoft-cloud/blog/2024/10/09/5-key-features-and-benefits-of-large-language-models/
Large Language Models (LLMs) with Google AI, accessed April 3, 2025, https://cloud.google.com/ai/llms
Practical Guide for LLMs in the Financial Industry | Automation Ahead Series, accessed April 3, 2025, https://rpc.cfainstitute.org/research/the-automation-ahead-content-series/practical-guide-for-llms-in-the-financial-industry
Leveraging Large Language Models (LLMs) in Finance: Development and Data Training, accessed April 3, 2025, https://www.pacificdataintegrators.com/blogs/llms-in-finance-development-and-data-training
LLM in Banking and Finance: Key Use Cases, Examples, and a Practical Guide – Shaip, accessed April 3, 2025, https://weareshaip.medium.com/llm-in-banking-and-finance-key-use-cases-examples-and-a-practical-guide-bd537e5d025c
LLMs: use cases in financial services – Nortal, accessed April 3, 2025, https://nortal.com/insights/llms-use-cases-in-financial-services/
What’s the future of LLM in finance? – Geniusee, accessed April 3, 2025, https://geniusee.com/single-blog/llm-in-finance
What are the pros and cons of using LLMs in compliance? – Global Relay, accessed April 3, 2025, https://www.globalrelay.com/resources/the-compliance-hub/compliance-insights/what-are-the-pros-and-cons-of-using-llms-in-compliance/
Maximizing compliance: Integrating gen AI into the financial regulatory framework | IBM, accessed April 3, 2025, https://www.ibm.com/think/insights/maximizing-compliance-integrating-gen-ai-into-the-financial-regulatory-framework
Knowledge Graph and LLM Integration: Benefits & Challenges – FalkorDB, accessed April 3, 2025, https://www.falkordb.com/blog/knowledge-graph-llm/
Knowledge Graphs & LLMs: Multi-Hop Question Answering – Neo4j, accessed April 3, 2025, https://neo4j.com/blog/developer/knowledge-graphs-llms-multi-hop-question-answering/
What is GraphRAG? An In-Depth Look at This Graph-Based Tool | by ODSC, accessed April 3, 2025, https://odsc.medium.com/what-is-graphrag-an-in-depth-look-at-this-graph-based-tool-29f1c86ca33d
Improving Retrieval Augmented Generation accuracy with GraphRAG – AWS, accessed April 3, 2025, https://aws.amazon.com/blogs/machine-learning/improving-retrieval-augmented-generation-accuracy-with-graphrag/
Unlocking Insights: GraphRAG & Standard RAG in Financial Services, accessed April 3, 2025, https://techcommunity.microsoft.com/blog/azure-ai-services-blog/unlocking-insights-graphrag–standard-rag-in-financial-services/4253311
GraphRAG Analysis for Financial Narrative Summarization and A Framework for Optimizing Domain Adaptation – ACL Anthology, accessed April 3, 2025, https://aclanthology.org/2025.finnlp-1.2/
GraphRAG Analysis for Financial Narrative Summarization and A Framework for Optimizing Domain Adaptation – ACL Anthology, accessed April 3, 2025, https://aclanthology.org/2025.finnlp-1.2.pdf
How to Build a Knowledge Graph – NebulaGraph, accessed April 3, 2025, https://www.nebula-graph.io/posts/how-to-build-knowledge-graph
How to build a knowledge graph in 9 simple steps – Lettria, accessed April 3, 2025, https://www.lettria.com/blogpost/how-to-build-a-knowledge-graph-in-9-simple-steps
What is Data Extraction? Definition, How it Works, and Techniques – Astera Software, accessed April 3, 2025, https://www.astera.com/type/blog/what-is-data-extraction-a-brief-guide/
Data Warehousing and Data Mining 101 – Panoply, accessed April 3, 2025, https://panoply.io/data-warehouse-guide/data-warehousing-and-data-mining-101/
How do you populate a knowledge graph? – Milvus, accessed April 3, 2025, https://milvus.io/ai-quick-reference/how-do-you-populate-a-knowledge-graph
How to Build a Knowledge Graph in 7 Steps – Neo4j, accessed April 3, 2025, https://neo4j.com/blog/knowledge-graph/how-to-build-knowledge-graph/
Constructing Knowledge Graphs From Unstructured Text Using LLMs – Neo4j, accessed April 3, 2025, https://neo4j.com/blog/developer/construct-knowledge-graphs-unstructured-text/
Dual Approaches to Building Knowledge Graphs: Traditional Techniques or LLMs, accessed April 3, 2025, https://bluetickconsultants.medium.com/dual-approaches-to-building-knowledge-graphs-traditional-techniques-or-llms-400fee0f5ac9
Building a Knowledge Graph: A Step-by-Step Guide for Beginners – SmythOS, accessed April 3, 2025, https://smythos.com/ai-agents/ai-tutorials/building-a-knowledge-graph/
How to Build a Knowledge Graph: A Step-by-Step Guide – FalkorDB, accessed April 3, 2025, https://www.falkordb.com/blog/how-to-build-a-knowledge-graph/
7 Best Open Source Graph Databases – PuppyGraph, accessed April 3, 2025, https://www.puppygraph.com/blog/open-source-graph-databases
Understanding Knowledge Graphs: Representation, Benefits, and Components – Lyzr AI, accessed April 3, 2025, https://www.lyzr.ai/glossaries/knowledge-graphs/
Ontology in Graph Models and Knowledge Graphs, accessed April 3, 2025, https://graph.build/resources/ontology
Ontology Design Best Practices – Part I – Enterprise Knowledge, accessed April 3, 2025, https://enterprise-knowledge.com/ontology-design-best-practices-part/
FIBO – OKG – EDM Council, accessed April 3, 2025, https://spec.edmcouncil.org/fibo/
FIBO – Finance Industry – EDM Council, accessed April 3, 2025, https://edmcouncil.org/frameworks/industry-models/fibo/
Reference Ontologies – Financial Regulation Ontology, accessed April 3, 2025, https://finregont.com/reference-ontologies/
Financial Information Business Ontology (FIBO):, accessed April 3, 2025, https://globalfintechseries.com/featured/financial-information-business-ontology-fibo-architecture-use-cases-and-implementation-challenges/
FIBO in Context | Ontotext, accessed April 3, 2025, https://www.ontotext.com/blog/fibo-in-context/
Finance Domain ontology transformed into an Enterprise Data Model., accessed April 3, 2025, https://fib-dm.com/finance-ontology-transform-data-model/
globalfintechseries.com, accessed April 3, 2025, https://globalfintechseries.com/featured/financial-information-business-ontology-fibo-architecture-use-cases-and-implementation-challenges/#:~:text=FIBO’s%20architecture%20consists%20of%20several,securities%2C%20loans%2C%20and%20derivatives.
How to use FIBO Viewer – OKG, accessed April 3, 2025, https://spec.edmcouncil.org/fibo/ontology
Financial Sector Domain Task Force | Object Management Group, accessed April 3, 2025, https://www.omg.org/fdtf/
edmcouncil/fibo: The Financial Industry Business Ontology (FIBO) defines the sets of things that are of interest in financial business applications and the ways that those things can relate to one another. In this way, FIBO can give meaning to any data (e.g., spreadsheets, relational databases, XML documents) that describe – GitHub, accessed April 3, 2025, https://github.com/edmcouncil/fibo
5 best practices to create powerful ontologies with LLMs – Lettria, accessed April 3, 2025, https://www.lettria.com/blogpost/5-best-practices-to-create-powerful-ontologies-with-llms
Ontology Development: How to Create Your First Ontology – Lettria, accessed April 3, 2025, https://www.lettria.com/blogpost/ontology-development-how-to-create-your-first-ontology
Top Use Cases for Graph Databases – Dgraph Blog, accessed April 3, 2025, https://dgraph.io/blog/post/use-case-graph-database/
Recognizing the Power of Graph Databases and Knowledge Graphs, accessed April 3, 2025, https://www.dbta.com/Editorial/Trends-and-Applications/Recognizing-the-Power-of-Graph-Databases-and-Knowledge-Graphs-166816.aspx
Knowledge Graph Extraction and Challenges – Graph Database & Analytics – Neo4j, accessed April 3, 2025, https://neo4j.com/blog/developer/knowledge-graph-extraction-challenges/
Knowledge Graph – Graph Database & Analytics – Neo4j, accessed April 3, 2025, https://neo4j.com/use-cases/knowledge-graph/
Elevate Fraud Detection With Neo4j on AWS: Uncover Hidden Patterns and Enhance Accuracy, accessed April 3, 2025, https://neo4j.com/blog/fraud-detection/neo4j-aws-fraud-detection/
New Amazon Bedrock capabilities enhance data processing and retrieval | AWS News Blog, accessed April 3, 2025, https://aws.amazon.com/blogs/aws/new-amazon-bedrock-capabilities-enhance-data-processing-and-retrieval/
Knowledge Graphs – Amazon Neptune – Amazon Web Services, accessed April 3, 2025, https://aws.amazon.com/neptune/knowledge-graphs-on-aws/
Popular and unique Knowledge Graph Use Cases for data analytics – ScikIQ, accessed April 3, 2025, https://scikiq.com/blog/popular-and-unique-knowledge-graph-use-cases-for-data-analytics/
2025’s Knowledge Graph Software You Can Trust, accessed April 3, 2025, https://www.pageon.ai/blog/knowledge-graph-software
Natural Language Querying of GraphDB in LangChain – Ontotext, accessed April 3, 2025, https://www.ontotext.com/blog/natural-language-querying-of-graphdb-in-langchain/
Ontotext Knowledge Graph, accessed April 3, 2025, https://www.ontotext.com/ontotext-knowledge-graph/
Ontotext GraphDB, accessed April 3, 2025, https://www.ontotext.com/products/graphdb/
What is Ontotext GraphDB – YouTube, accessed April 3, 2025, https://www.youtube.com/watch?v=6Nc9Yy1ttWs
Graph RAG Tutorial for Financial Services – Ontotext, accessed April 3, 2025, https://www.ontotext.com/graph-rag-tutorial/
KGF 2021 Talks: Knowledge Graphs in the Financial Services Industry | Ontotext, accessed April 3, 2025, https://www.ontotext.com/knowledgehub/videos/kgf21-talks-knowledge-graphs-in-the-financial-services-industry/
Crafting a Knowledge Graph: The Semantic Data Modeling Way – Ontotext, accessed April 3, 2025, https://www.ontotext.com/blog/knowledge-graph-with-semantic-data-modeling/
Solutions for Financial Services – Ontotext, accessed April 3, 2025, https://www.ontotext.com/solutions/financial-services/
Best practices for maintaining and updating a knowledge graph, accessed April 3, 2025, https://knowledgegraph.dev/article/Best_practices_for_maintaining_and_updating_a_knowledge_graph.html
A Beginner’s Guide to Knowledge Graph Optimization in 2025 – PingCAP, accessed April 3, 2025, https://www.pingcap.com/article/knowledge-graph-optimization-guide-2025/
Insights, Techniques, and Evaluation for LLM-Driven Knowledge Graphs | NVIDIA Technical Blog, accessed April 3, 2025, https://developer.nvidia.com/blog/insights-techniques-and-evaluation-for-llm-driven-knowledge-graphs/
Semantic RAG: Enhancing LLMs with Knowledge Graphs, accessed April 3, 2025, https://www.poolparty.biz/blogposts/semantic-rag-knowledge-graphs-llms
How to Implement Knowledge Graphs and Large Language Models (LLMs) at the Enterprise Level | OTAKOYI, accessed April 3, 2025, https://otakoyi.software/blog/how-to-implement-knowledge-graphs-and-large-language-models-llms-at-the-enterprise-level
Knowledge Graph Query Engine – LlamaIndex, accessed April 3, 2025, https://docs.llamaindex.ai/en/stable/examples/query_engine/knowledge_graph_query_engine/
[2410.06062] LLM-based SPARQL Query Generation from Natural Language over Federated Knowledge Graphs – arXiv, accessed April 3, 2025, https://arxiv.org/abs/2410.06062
Building Knowledge Graph Agents With LlamaIndex Workflows – Neo4j, accessed April 3, 2025, https://neo4j.com/blog/knowledge-graph/knowledge-graph-agents-llamaindex/
Text2Cypher: Bridging Natural Language and Graph Databases – arXiv, accessed April 3, 2025, https://arxiv.org/html/2412.10064v1
Introducing the Neo4j Text2Cypher (2024) Dataset, accessed April 3, 2025, https://neo4j.com/blog/developer/introducing-neo4j-text2cypher-dataset/
(PDF) Text2Cypher: Bridging Natural Language and Graph Databases – ResearchGate, accessed April 3, 2025, https://www.researchgate.net/publication/387078584_Text2Cypher_Bridging_Natural_Language_and_Graph_Databases
Text2Cypher: Bridging Natural Language and Graph Databases | PromptLayer, accessed April 3, 2025, https://www.promptlayer.com/research-papers/ask-your-database-in-plain-english
What Are Knowledge Graph Embeddings? – Ontotext, accessed April 3, 2025, https://www.ontotext.com/knowledgehub/fundamentals/what-are-knowledge-graph-embeddings/
Machine Learning on Graphs: Knowledge Graphs and Embeddings – Mosaic Data Science, accessed April 3, 2025, https://mosaicdatascience.com/2023/02/15/machine-learning-on-graphs-knowledge-graphs-and-embeddings/
Multidimensional Knowledge Graph Embeddings for International Trade Flow Analysis, accessed April 3, 2025, https://arxiv.org/html/2410.19835v1
What are graph embeddings ? – NebulaGraph, accessed April 3, 2025, https://www.nebula-graph.io/posts/graph-embeddings
Graph LLMs The Next AI Frontier in Banking and Insurance Transformation. – PwC, accessed April 3, 2025, https://www.pwc.com/m1/en/publications/documents/2024/graph-llms-the-next-ai-frontier-in-banking-and-insurance-transformation.pdf
Supercharging Fraud Detection in Financial Services with Graph Neural Networks | NVIDIA Technical Blog, accessed April 3, 2025, https://developer.nvidia.com/blog/supercharging-fraud-detection-in-financial-services-with-graph-neural-networks/
Financial Fraud Detection using Jump-Attentive Graph Neural Networks – arXiv, accessed April 3, 2025, https://arxiv.org/html/2411.05857v1
Graph Neural Networks: Fraud Detection and Protein Function Prediction, accessed April 3, 2025, https://towardsdatascience.com/graph-neural-networks-fraud-detection-and-protein-function-prediction-08f9531c98de/
Supercharging Fraud Detection in Financial Services with Graph Neural Networks – NVIDIA, accessed April 3, 2025, https://resources.nvidia.com/en-us-financial-services-industry/supercharging-fraud-detection-in-financial-services-with-graph-neural-networks
[2502.19305] Corporate Fraud Detection in Rich-yet-Noisy Financial Graph – arXiv, accessed April 3, 2025, https://www.arxiv.org/abs/2502.19305
Embeddings + Knowledge Graphs: The Future of Generative AI | by Chris Galleta | Medium, accessed April 3, 2025, https://medium.com/@thcookieh/embeddings-knowledge-graphs-the-future-of-generative-ai-9abf70bdee19
Research on the Application of Knowledge Graphs in Bank Risk Management, accessed April 3, 2025, https://www.ewadirect.com/proceedings/aemps/article/view/17745
LLM Security 101: Protecting Large Language Models from Cyber Threats, accessed April 3, 2025, https://blog.qualys.com/misc/2025/02/07/llm-security-101-protecting-large-language-models-from-cyber-threats
What are the Top Security Risks of Using Large Language Models (LLMs)? – Metomic, accessed April 3, 2025, https://www.metomic.io/resource-centre/what-are-the-top-security-risks-of-using-large-language-models-llms
Large Loss of Money? Choose Your LLM Security Solution Wisely. – Akamai, accessed April 3, 2025, https://www.akamai.com/blog/security/2024-november-llm-security-financial-impact
(PDF) Trust and Privacy in Knowledge Graphs – ResearchGate, accessed April 3, 2025, https://www.researchgate.net/publication/331888241_Trust_and_Privacy_in_Knowledge_Graphs
How Knowledge Graph Enhances Data Privacy Regulations – BytePlus, accessed April 3, 2025, https://www.byteplus.com/en/topic/488937
How To Preserve Data Privacy In LLMs In 2025 – Protecto.ai, accessed April 3, 2025, https://www.protecto.ai/blog/how-to-preserve-data-privacy-in-llms
Data Privacy and Compliance for Large Language Models (LLMs) | by Sanjay K Mohindroo, accessed April 3, 2025, https://medium.com/@sanjay.mohindroo66/data-privacy-and-compliance-for-large-language-models-llms-37d8179ac12b
Knowledge Graphs: Access, Manage, Share, and Derive Insights From Data Like Never Before – BurstIQ, accessed April 3, 2025, https://burstiq.com/knowledge-graphs-access-manage-share-and-derive-insights-from-data-like-never-before/
Privacy-Preserving Large Language Models: Mechanisms, Applications, and Future Directions – arXiv, accessed April 3, 2025, https://arxiv.org/html/2412.06113v1
Managing Risks of LLMs in Finance – Lumenova AI, accessed April 3, 2025, https://www.lumenova.ai/blog/managing-risks-large-language-models-financial-services/

Balancing Innovation and Risk: Current and Future Use of LLMs in the Financial Industry, accessed April 3, 2025, https://insideainews.com/2025/02/06/balancing-innovation-and-risk-current-and-future-use-of-llms-in-the-financial-industry/