Creating a Semantic Data Model with Termboard

Termboard is specifically designed and highly optimized for creating semantic data models. Indeed, its name, "Termboard," reflects its core purpose: providing a dedicated canvas for organizing and connecting the fundamental components (terms) of your semantic models.

Termboard Functionalities

Termboard offers a comprehensive suite of features to assist you throughout the semantic data modeling process. Below a short overview or you can watch the videos on youtube, for example How to create a knowledge graph

• Efficient Data Entry via Bulk Imports:

  • Streamline the population of your model by importing terms, definitions, and relations in bulk.
  • See: Bulk Add Terms and Relations

• Automated Relation Inference (Inheritance):

  • Leverage Termboard's ability to infer (inherit) relations, significantly reducing manual effort and ensuring consistency across your model.
  • See: Relation Inheritance

• AI-Powered Assistance:

  • Utilize integrated AI support to accelerate your modeling:
    • Discover Parents/Children: Get suggestions for hierarchical relationships.
    • Obtain Definitions: Quickly find or refine definitions for your terms.
    • Find Related Terms: Identify potential connections within your domain.
  • See: Term menu

• Intuitive Graphical Editing and Styling:

  • Graphical Editing: Visually create, modify, and manage terms and relations directly on the board.
  • Grouping and Styling: Organize and highlight terms and relations through customizable grouping and styling options.
  • Conditional Formatting: Apply visual formatting to terms based on the values of their custom fields, allowing for quick identification of specific data patterns.
  • See: Canvas, Extra Fields

• Interactive Graph Exploration:

  • Navigate and understand even the largest and most complex semantic graphs with interactive exploration tools.
  • See: Interactive Graph

• Pathfinding Analysis:

  • Quickly determine the shortest path between any two terms in your model, revealing hidden connections and relationships.
  • See: Find path

• Flexible Import and Export Options:

  • Interoperate with other systems and tools through a variety of import and export formats, including:
    • Semantic Web Standards: OWL (Web Ontology Language), JSON-LD (JSON for Linking Data).
    • Generic Data Formats: JSON, Excel.
    • Documentation & Presentation: PDF, PowerPoint, and various image formats.
    • See: Import/Export

• Diverse Model Views and Editors:

  • Taxonomy View: Gain a focused hierarchical perspective of your terms.
  • Relation Types Editor: Define and manage the types of relationships in your model.
  • Table Editor: Work with your terms and relations in a structured, tabular format for efficient editing.
  • See Taxonomy view, Relation types, Table Editor

• LLM Prompts for Model Generation:

  • Jumpstart your modeling process by using Large Language Model (LLM) prompts to generate initial semantic data models from free-form text descriptions or specified concepts.
  • See Prompts

• Comprehensive Semantic Checks:

  • Ensure the quality and consistency of your model with built-in semantic checks that identify common issues, such as:
    • Incorrect naming conventions (e.g., plurals in singular names).
    • Duplicate terms or relations.
    • Missing descriptions.
    • Parents mentioned within child descriptions.
    • Multiple parents for terms in a strict hierarchy.
    • Circular hierarchies.
    • Other custom validation rules.
  • See: Semantic Checks

• Cardinality Definition in Relations:

  • Define cardinalities for your relations (e.g., one-to-one, one-to-many) to specify the number of allowed instances between related terms, enhancing the precision of your model.
  • See: Cardinalities

Semantic modeling

A Semantic Data Model (SDM) is a conceptual framework designed to represent data in a way that explicitly captures both its meaning (semantics) and the complex relationships inherent within it.

Unlike traditional data models (like relational models) that primarily focus on structure (tables and columns), an SDM emphasizes how data elements relate to real-world concepts, events, and business entities. Its core goal is to provide a unified, business-friendly view of data that is easily understood by both humans and machines, often forming the foundation of Knowledge Graphs and Ontologies.

Term/Concept/Property

In semantic data modeling, concepts and attributes (or properties) are differentiated as distinct term types. This approach helps to structure and organize knowledge in a clear and logical manner.

Term

If you do not want to differentiate or are not sure yet which type your term should be in your semantic model, you can use "Term" as default value. It is also set as default when you create a new term.

Concept

Concepts represent the main entities or ideas within your domain. They are the "nouns" of your data model and can exist independently. A concept might be a person, a product, a location, or an event. They are typically organized hierarchically, allowing you to establish relationships like "is-a" or "part-of."

INFO

In a data model for a university, "Student," "Course," and "Professor" would be concepts.

Property (=Attribute)

Attributes, or properties, describe or define a concept. They are the properties or qualities of an entity, representing the "adjectives" of your data model. Attributes cannot exist on their own; they are always associated with a specific concept.

INFO

For the concept "Student," attributes might include "Student ID," "GPA," and "Major." These properties provide specific details about each individual student.

Differentiating with Term Type

By assigning different term types (e.g., "Concept" and "Property") to your data, you create a more structured and formal semantic model. This differentiation allows for:

• Clarity: It makes the model easier to understand, as it clearly distinguishes between what an entity is and what describes it.

• Consistency: It ensures that all data related to a concept is consistently defined and structured.

• Reasoning: It facilitates automated reasoning and analysis, as systems can understand the relationships between different types of terms and their properties.

This differentiation is a core principle in creating a robust semantic data model, such as those used in knowledge graphs and ontologies.

Benefits of Semantic Data Modeling

1. Gen AI (LLMs)

This is one of the most significant modern advantages. Large Language Models (LLMs) and Generative AI agents are powerful, but they are prone to hallucinations (generating confident but false information) when dealing with private, structured enterprise data.

LLM Problem	SDM Solution
Hallucinations	The SDM acts as a grounding layer. It constrains the LLM's output to factual relationships, metrics, and definitions that are verified and governed by the business.
Inconsistent Answers	When an LLM translates a natural language question (e.g., "What was the Q2 profit?") into a technical query, the SDM ensures the LLM is using the single, standardized formula for "Q2 profit," regardless of the underlying database structure.
Difficulty with Context	The SDM provides explicit business context for every piece of data. The LLM understands that an entity called "Customer" relates to "Order," and that a "Product Category" is a dimension of a "Sale," allowing it to generate far more accurate and relevant data-driven answers.

By sitting between the user and the raw data, the semantic model ensures that the AI is answering business questions, not just performing database lookups.

---

2. Data Governance

Governance is impossible without consistent definitions. SDMs enforce control by centralizing the "what" and "how" of data consumption.

• Centralized Definitions: SDMs move crucial business logic—like the calculation for Churn Rate or Total Revenue—out of individual BI reports and into a single, governed place. When the definition changes, it changes once for everyone.
• Compliance & Auditing: Since the SDM formally documents the meaning and lineage of data in business terms, it drastically simplifies compliance reporting (e.g., GDPR, SOX). Auditors can see the business context for data, not just raw table structures.
• Data Security: Access rules can be applied at the semantic layer (e.g., "Sales managers can only see data for their region"), abstracting the complexity of securing every single underlying database table.

---

3. Data Integration and Interoperability

In modern, decentralized data architectures (like Data Mesh), SDMs are vital for allowing different systems to talk to each other without constant, custom ETL (Extract, Transform, Load) work.

• Single View: An SDM allows you to create a unified conceptual view of an entity (e.g., "Employee") even if its data is physically spread across HR, Finance, and Payroll systems. It maps the meaning together, eliminating data silos at the consumption layer.
• Foundation for Knowledge Graphs: For highly complex domains, semantic models are the conceptual schema for building a Knowledge Graph. This graph-based structure explicitly models the relationships between entities, enabling sophisticated network analysis and inference that is impossible with traditional relational models.
• Decentralized Data Ownership: SDMs allow different domain teams to own their data while ensuring that all their outputs conform to a shared, enterprise-wide standard set of business terms, promoting consistency across the organization.

4. Software Development

An SDM serves as the definitive blueprint for any application that interacts with the business's core data, streamlining the development process in several key ways:

• Accelerated Development and Design:

  • Ready-Made Logic: Developers don't waste time figuring out how to model core business entities. The SDM provides a tested, pre-approved conceptual structure for objects like Customer, Invoice, and Product.
  • API Design Foundation: The model's entities and relationships are often translated directly into clean, intuitive APIs or microservices (the way applications talk to each other), significantly speeding up the backend coding phase.

• Improved Quality and Reduced Bugs:

  • Eliminating Ambiguity: By enforcing a single source of truth for business terminology, the SDM ensures that every developer and application uses the exact same logic and definition for every metric or relationship. This drastically reduces bugs caused by conflicting assumptions.
  • Validation and Constraints: The semantic model explicitly defines constraints and rules (e.g., "an Order must belong to a Customer"). These rules can be enforced at the application layer, ensuring only high-quality, valid data enters the system.

• Enhanced Maintainability and Decoupling:

  • Insulation from Change: The SDM acts as a stable buffer between the application logic and the underlying technical data storage (the physical database). If the database technology or schema changes (e.g., swapping SQL tables for a graph database), the application layer consuming the semantic model often requires minimal to no change. This decoupling is key to future-proofing your applications and reducing long-term maintenance costs.
  • Onboarding: New developers can quickly understand the business domain by reviewing the SDM, rather than having to reverse-engineer complex database tables and thousands of lines of code.

In essence, the semantic data model provides the specification for the business domain, allowing developers to focus on the user experience and technical performance, rather than reinventing the core business logic.

Modeling Workshop

This guide walks you through the 10 steps to organize a successful semantic data modeling workshop with subject matter experts.

1: Domain Scope

Activity: Meet with the designated domain lead or a senior business expert to clearly define the scope of the workshop.

Output:

• Domain Boundaries: Confirm the exact boundaries of the selected business domain (e.g., "Finance" - specifically focusing on "Financial Reporting").
• Workshop Objectives: Agree on the primary objectives of the workshop (e.g., "To create a high-level semantic model for the data/business domain, focusing on key financial reporting concepts and their relationships").
• Participant Identification: Identify key business experts from different areas within the domain who possess a broad understanding of data and business processes. Aim for a diverse group representing different perspectives.

TIP

The domain scope could be provided to an LLM (ChatGPT etc) later to pre-generate a very initial list of potential concepts as a starting point if needed, but this should be used cautiously to avoid biasing the business experts' input.
An example prompt here. The output can be pasted in Termboard using Tools > Add in Bulk/Chatbot. See here for the output using Gemini 2.5

2: Workshop Preparation

Activity: Prepare all necessary materials for the workshop.

Materials:

• Workshop Agenda: A clear agenda outlining the workshop steps and expected outcomes.
• Domain Overview (Optional): A brief, high-level document summarizing the data domain's purpose and key business functions (useful for setting context).
• Examples of Semantic Models (Simple): Illustrative examples of simple semantic models from other domains (avoid showing examples from the target domain to prevent pre-conceived ideas). Focus on visual representations of concepts and relationships.
• Concept & Relationship Templates (Physical or Digital): Prepare templates (e.g., sticky notes, digital whiteboard templates) for participants to easily capture concepts, descriptions, and relationships. Pre-printed concept types like "Event", "Party", "Product" can be helpful starting points.
• LLM Access (Prepared): Ensure access to an LLM tool (via browser, API, or dedicated platform) for use during specific workshop steps. Have pre-defined prompts ready to guide the LLM in concept extraction and description refinement.

TIP

(more info soon...)

3: Kick-off & Context

Activity:

• Welcome & Introductions: Facilitator welcomes participants and introductions are made.
• Workshop Objectives & Agenda Review: Clearly state the workshop's objectives and review the agenda. Emphasize the collaborative nature of the workshop and the importance of business expertise.
• Domain Context Setting: Briefly reiterate the defined domain scope and its importance to the bank. Use the optional domain overview document if prepared.
• Semantic Modeling 101 (Very Brief): Provide a very brief and non-technical introduction to semantic modeling. Explain:
  • Concepts: "Business things we talk about" (entities, events, attributes).
  • Relationships: "How these things are connected."
  • Hierarchy (is-a): "Different types of things."
  • Goal: "To create a shared understanding of the data in your domain."
• Key Message: "Forget about technical details for now, focus on the business language you use every day."

4: Usecases Identification

Activity: Engage business experts in identifying and listing key data usecases within their domain.

Facilitation:

• Guidance Question: "Think about the reports, analyses, data feeds, and outputs your domain produces for other domains and uses itself. What are the key 'things' you deliver or work with as data?"
• Brainstorming Session: Use a whiteboard or digital tool to collect data usecase names and brief descriptions. Encourage everyone to contribute.
• Focus on Business Value: Emphasize the business purpose of each usecase. "Why is this report important? What business question does it answer?"

Output: A list of relevant usecases with business-oriented descriptions.

TIP

Potentially, if the domain is very broad, you could ask an LLM (before the workshop) to generate a sample list of data product types for the "Finance" domain (e.g., "Financial Reporting data products"). However, it's generally better to let the business experts drive this step organically.

5: Concept Extraction

Activity: Analyze the usecase for data product names and descriptions to extract initial core business concepts.

Facilitation:

• Guidance Question: "Looking at the data product names and descriptions, what are the key 'nouns' or 'business entities' being referred to?"
• Concept Identification: For each data product, ask: "What are the key things this data product is about?" Write down these concepts on sticky notes or digital templates.
• Grouping & Consolidation: Group similar concepts and consolidate duplicates. Discuss and agree on the most appropriate names for the core concepts.
• Visualise the concepts: Write them on a whiteboard or enter the concepts in Termboard

Output: A canvas with 10-20 core concepts representing key business entities, events, or attributes.

TIP

• Concept Suggestion: Feed the list of usecases or data product descriptions (from Step 4) into an LLM. Prompt the LLM with: "Identify the key business entities, concepts, or things mentioned in these descriptions. List them as single words or short phrases." The LLM can suggest potential concepts that might have been missed by the group.
• Concept Description Refinement (Later in the step): Once the business experts have a list of core concepts, you can ask the LLM to generate initial draft descriptions for each concept based on the data product descriptions where they appear. This provides a starting point for business experts to refine and validate. Prompt: "Provide a brief business-oriented description for the concept '{Concept Name}', considering its use in the context of these data product descriptions: [List relevant data product descriptions]."

6: Hierarchy Building

Activity: Build the "is-a" hierarchy for the core concepts and identify superclasses.

Facilitation:

• Guidance Question: "For each core concept, ask: 'What kind of [Concept] is this?'" (e.g., "What kind of 'Financial Report' is 'Profit & Loss Statement'?").
• Iterative Superclass Identification: Start with the core concepts and work upwards. "Is there a more general category that includes this concept?" Continue this process to reach higher-level, more abstract concepts.
• Stop at Relevant Abstraction: Stop adding superclasses when you reach concepts that are broadly relevant across domains (e.g., 'Agreement', 'Event', 'Product', 'Document', 'Record', 'Classification'). Avoid overly generic terms unless structurally necessary.
• Visual Hierarchy: Visually arrange the concepts in a hierarchical structure (e.g., using a tree diagram on a whiteboard or Termboard).

Output: A hierarchical structure of concepts with "is-a" relationships + expanded list of concepts adding the superclasses.

TIP

Superclass Suggestion: For a given concept, you can ask the LLM: "Suggest potential superclasses (more general categories) for the concept '{Concept Name}' within the context of financial reporting/the Finance domain." The LLM might suggest broader categories based on its knowledge base.

However, business expert validation is crucial here as LLM suggestions might not always be business-relevant.

7: Concept Definitions

Activity: Define and refine the descriptions for all concepts (core and superclasses).

Facilitation:

• Guidance Question: "For each concept, ensure we have a clear, concise, and business-oriented description. What does this concept mean in our business context?"
• Collaborative Definition: Work with the business experts to define each concept in their own words. Focus on clarity and shared understanding.
• Consistency Check: Ensure descriptions are consistent and avoid jargon or overly technical terms.

Output: Complete descriptions for all concepts in a document or using Termboard. Use an LLM to review the definitions for clarity and consistency

TIP

Improve Definitions: In the Term sidebar of Termboard, select the icon with the stars, then select Improve Definition and paste the clipboard in an LLM to get an improved definitions for a concept

8: Relations Identification

Activity: Identify meaningful relationships other than "is-a" between the concepts.

Facilitation:

• Guidance Question: "Now, think about how these concepts interact. What are the key business relationships between them?"
• Relationship Brainstorming: Ask questions like: "What actions or processes connect these concepts?", "Does one concept influence or depend on another?", "Does one concept contain or manage another?"
• Relationship Types: Suggest relationship types to get started (e.g., 'reports on', 'manages', 'uses', 'is derived from', 'applies to', 'contains', 'is denominated in').
• Relationship Capture: Document the relationships using source concept, relationship name, and target concept.

Output: List of general relationships between concepts.

TIP

Relationship Type Suggestion: For a pair of concepts, you could ask the LLM: "Suggest potential business relationships between '{Concept A}' and '{Concept B}' in the context of financial reporting." However, rely more heavily on the business experts for this step, as understanding the nuances of business relationships is critical. LLM suggestions might be too generic or miss domain-specific relationships.

9: Review & Validation

Activity: Review the complete semantic model (concepts, hierarchies, and relationships) with the business experts.

Facilitation:

• Guidance Question: "Let's review what we've created. Does this model accurately represent the key concepts and relationships in your domain?"
• Validation & Refinement: Ask for feedback, corrections, and refinements. Ensure everyone agrees that the model is a reasonable initial representation.
• Business Sign-off (Informal): Aim for informal agreement and sign-off from the business experts that the model is a good starting point.

Output: Validated initial semantic model.

10: Documentation & Iteration

Activity:

• Document the Model: Formalize the semantic model (create digital diagrams, populate tables as per the requested format, etc.).
• Share & Communicate: Share the initial model with the workshop participants and the broader domain team for feedback and awareness.
• Iteration Planning: Emphasize that this is an initial model and will likely evolve. Plan for future iterations and workshops to refine and expand the model as needed.

TIP

Documentation Generation: LLMs might assist in generating documentation from the structured model data (e.g., generating concept glossaries, relationship descriptions).