Enhancement Request: Enable Full Utilization of node_properties and relationship_properties in LLMGraphTransformer

[dhiaaeddine16]

Dear Maintainers,

I would like to propose an enhancement to the llm-graph-builder project to fully leverage the node_properties and relationship_properties parameters in the LLMGraphTransformer. While these parameters are defined during initialization, they are not currently utilized to extract detailed properties for nodes and relationships in the generated knowledge graph.

Proposed Enhancement

Enable the use of node_properties and relationship_properties to allow users to specify the attributes they want to extract, providing greater control and flexibility in generating knowledge graphs.

Key Benefits

1. Enhanced Graph Precision: Extract richer, more detailed data for nodes and relationships.
2. Customization: Allow users to tailor property extraction to specific domain requirements.

I am confident that this enhancement will greatly improve the functionality of the llm-graph-builder and help make it even more valuable to the community. I am also happy to contribute to the implementation of this feature and assist in any way possible to help the project grow.

Thank you for considering this request, and I look forward to any feedback.

Best regards,
Dhiaa eddine.

[dhiaaeddine16]

I’m interested in contributing to this enhancement. Could you please provide guidance on how I can contribute to the implementation of this feature?

[kaustubh-darekar]

Hi @dhiaaeddine16 thanks for suggestions.

currently we have option to pass allowed nodes and allowed relationships through UI.
in next release we are also adding additional instructions to be passed to LLMGraphTransformer refer(langchain-ai/langchain-experimental#23).

We can also make node_properties and relationship properties configurable by passing those through UI (Need to be implemented)

[dhiaaeddine16]

Hi @kaustubh-darekar,

Thank you for the update and acknowledgment! I appreciate the ongoing improvements to the project.

The addition of the option to pass additional instructions to the LLMGraphTransformer in the next release sounds promising, and it aligns well with my suggestion for increased flexibility.

I also fully support the idea of making node_properties and relationship_properties configurable via the UI. This feature would significantly enhance usability by allowing users to tailor the graph-building process to their specific needs more intuitively.

If there's anything I can contribute or test regarding these enhancements, please let me know. Excited to see this project evolve further!

[dhiaaeddine16]

Hello @jexp, @tomasonjo, @kartikpersistent, and @kaustubh-darekar,

I hope you're all doing well. I would like to highlight the LLMGraphTransformer repository. I believe its capabilities could significantly enhance the LLM Graph Builder project by providing valuable improvements in functionality.

I would appreciate your expert insights on how best to integrate this solution into our project. Thank you for your time and consideration.

Best regards,

[tomasonjo]

Hi @dhiaaeddine16

can you please elaborate what are the enhancements compared to LangChain LLM Graph Transformer?

Thanks

[dhiaaeddine16]

Hi @tomasonjo,

Thank you for your interest in the LLMGraphTransformer. I'd like to highlight some key enhancements it offers:

• Comprehensive Property Extraction Across All LLMs:
  Unlike LangChain's implementation, which requires function calling capabilities for property extraction, LLMGraphTransformer supports node and relationship property extraction with any language model, even those lacking function calling features.

• Detailed Schema Definition:
  LLMGraphTransformer allows for precise schema definitions where each node and relationship type can have specific properties. This contrasts with LangChain's approach, which separates node types from their properties and relationship types from their properties, potentially leading to less cohesive schema representations.

I hope this clarifies the enhancements LLMGraphTransformer brings to the table. Please let me know if you have any further questions or need additional details.

Best regards,
Dhiaa-eddine

[tomasonjo]

LC version also supports extraction without function use. How are you implementing properties per node and does this approach scale to like 5 node types with 3 properties and 10 relationships?

[dhiaaeddine16]

What do you mean by the LC version.

[tomasonjo]

LangChain version

[dhiaaeddine16]

Hi @tomasonjo,

LC version does not support extraction of node_properties and relationship_properties with LLMs that don't support native function calling (LC code).

Additionally, the LC version separates node types from their properties and relationship types from their properties, potentially leading to less cohesive schema representations (LC code).

Best regards,
Dhiaa-eddine