# Custom Document Intelligence Model for Energy Certificates - Complex Table Relationships

Question

Use Case

We have energy certificate PDFs containing multiple interconnected tables that describe building components and their material compositions.

Example Table Structure

Main Components Table:

Component_IDDescriptionThickness [cm]U-Value [W/m²K]Area [m²]X01Roof30.00.1290.0Y01North Wall24.00.1566.0Z01Window Type 18.00.931.0### Detailed Material Breakdown Table (for X01): LayerMaterial DescriptionThickness [cm]λ [W/mK]Density [kg/m³]1Concrete Layer25.02.524002Vapor Barrier0.30.236503Insulation Panel18.00.038120## Goal We want to:

1. Extract both types of tables using Azure Document Intelligence
2. Maintain the relationship between component IDs (e.g., X01) and their material breakdowns
3. Build a solution that automatically links related tables post-extraction

Technical Questions

1. What's the best approach to train the model to recognize these relationship identifiers?
2. How can we structure the labels to capture both the tabular data and the linking information?
3. Are there any best practices for handling hierarchical table relationships in Document Intelligence?
4. What post-processing approaches would you recommend?

Answer 1

Hi Belkiss Souayed,
Greetings & Welcome to Microsoft Q&A forum! Thanks for posting your query!

To train a custom Document Intelligence model for energy certificates, start with a well-organized dataset containing various documents with tables and relationships. Use Azure Document Intelligence Studio to carefully label tables, marking component IDs and material breakdowns. Choose a custom template model for structured documents or a custom neural model for more flexible layouts.

For labeling, ensure consistency by defining fields clearly and using a unique identifier (e.g., Component_ID) to link related tables. Hierarchical labeling helps capture relationships between different document sections for better accuracy.

To handle hierarchical table relationships, analyze the document structure and segment it logically. Azure AI Document Intelligence supports hierarchical document structure analysis, making it easier to extract and process data in a structured format like JSON.

For post-processing, you can combine multiple models using composed models, format extracted data in markdown for readability, and automate table linking with scripts based on unique identifiers. This ensures relationships within the document are maintained accurately.tifiers, ensuring that relationships within the document are preserved post-extraction.

Hope this helps. Do let us know if you have any further queries.

---

If this answers your query, do click Accept Answer and Yes for was this answer helpful.