Add Support for Node and Relationship Properties in LLMGraphTransformer for LLMs without native function calling

[dhiaaeddine16]

This pull request addresses #40 by introducing support for the node_properties and relationship_properties parameters in the LLMGraphTransformer for LLMs without native function calling. Previously, the LLMGraphTransformer restricted the use of these parameters when working with LLMs lacking native function calling capabilities or when ignore_tool_usage was set to True. This limitation compelled users to choose between utilizing lightweight or open-source LLMs without structured property enforcement and opting for more costly or proprietary models that support function calling. The current update removes this restriction, enabling the use of node_properties and relationship_properties across a broader range of LLMs, thereby enhancing flexibility and functionality.

[dhiaaeddine16]

Hi @ccurme, @efriis, @Levara, @tomasonjo could someone please review this PR? 🙏
Let me know if any adjustments are needed!

[dhiaaeddine16]

Hi @kartikpersistent, @kaustubh-darekar

Following our discussion in Issue #1002, I’ve implemented support for node_properties and relationship_properties in the LLMGraphTransformer for LLMs without native function calling or when ignore_tool_usage was set to True.

[tomasonjo]

Which models have you tested this with? Looks a bit complicated for the LLM

[dhiaaeddine16]

I have tested it with Llama-3.3-70B-Instruct-Turbo.

[tomasonjo]

Can you try with some 7b models?

[dhiaaeddine16]

with this conf:

text = """
Marie Curie, born in 1867, was a Polish and naturalised-French physicist and chemist who conducted pioneering research on radioactivity.
She was the first woman to win a Nobel Prize, the first person to win a Nobel Prize twice, and the only person to win a Nobel Prize in two scientific fields.
Her husband, Pierre Curie, was a co-winner of her first Nobel Prize, making them the first-ever married couple to win the Nobel Prize and launching the Curie family legacy of five Nobel Prizes.
She was, in 1906, the first woman to become a professor at the University of Paris.
"""

model_name = "mistralai/Mistral-7B-Instruct-v0.2"
llowed_nodes = ["Award", "Person", "Country", "Organization"]
allowed_relationships = [
    ("Person", "SPOUSE", "Person"),
    ("Person", "NATIONALITY", "Country"),
    ("Person", "WORKED_AT", "Organization"),
    ("Person", "LOCATED_IN", "Country"),
    ("Person", "AWARDED", "Award"),
]
node_properties=["born_year"]
relationship_properties=["role"]
additional_instructions = """- Your goal is to identify and categorize entities while ensuring that specific data types such as dates, numbers, revenues, and other non-entity information are not extracted as separate entity. Instead, treat these as properties associated with the relevant entities."""
ignore_tool_usage = True
strict_mode = True

GraphDocument(nodes=[Node(id='Marie Curie', type='Person', properties={'born_year': '1867'}), Node(id='Poland', type='Country', properties={}), Node(id='France', type='Country', properties={}), Node(id='University of Paris', type='Organization', properties={}), Node(id='Nobel Prize', type='Award', properties={}), Node(id='Pierre Curie', type='Person', properties={})], relationships=[Relationship(source=Node(id='Marie Curie', type='Person', properties={}), target=Node(id='Poland', type='Country', properties={}), type='NATIONALITY', properties={}), Relationship(source=Node(id='Marie Curie', type='Person', properties={}), target=Node(id='France', type='Country', properties={}), type='NATIONALITY', properties={}), Relationship(source=Node(id='Marie Curie', type='Person', properties={}), target=Node(id='University of Paris', type='Organization', properties={}), type='WORKED_AT', properties={'role': 'professor'}), Relationship(source=Node(id='Marie Curie', type='Person', properties={}), target=Node(id='Nobel Prize', type='Award', properties={}), type='AWARDED', properties={}), Relationship(source=Node(id='Marie Curie', type='Person', properties={}), target=Node(id='Nobel Prize', type='Award', properties={}), type='AWARDED', properties={'role': 'co-winner'}), Relationship(source=Node(id='Pierre Curie', type='Person', properties={}), target=Node(id='Marie Curie', type='Person', properties={}), type='SPOUSE', properties={}), Relationship(source=Node(id='Marie Curie', type='Person', properties={}), target=Node(id='Nobel Prize', type='Award', properties={}), type='AWARDED', properties={'role': 'co-winner'})], source=Document(metadata={}, page_content='\nMarie Curie, born in 1867, was a Polish and naturalised-French physicist and chemist who conducted pioneering research on radioactivity.\nShe was the first woman to win a Nobel Prize, the first person to win a Nobel Prize twice, and the only person to win a Nobel Prize in two scientific fields.\nHer husband, Pierre Curie, was a co-winner of her first Nobel Prize, making them the first-ever married couple to win the Nobel Prize and launching the Curie family legacy of five Nobel Prizes.\nShe was, in 1906, the first woman to become a professor at the University of Paris.\n'))

[dhiaaeddine16]

If issues occur with small models, we can separate the two prompts—the new one and the old one—and create two modes: advanced and simple.

[tomasonjo]

Can you make it so that the prompt only changes if someone defines node or rel props, otherwise it stays the same, so we dont introduce breaking changes V ned., 9. feb. 2025, 11:47 je oseba Dhiaa Eddine ***@***.***> napisala:

…

If issues occur with small models, we can separate the two prompts—the new one and the old one—and create two modes: advanced and simple. — Reply to this email directly, view it on GitHub <#42 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEYGGTMDKNOL4XKI2PJKLB32O4W5LAVCNFSM6AAAAABWO536RKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNBWGE3TENBVGI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[dhiaaeddine16]

Hi @tomasonjo, thanks for your valuable input. Based on your suggestion, I've updated the implementation so that the prompt is modified only when node_properties or relationship_properties are explicitly provided. This ensures that if these parameters aren’t defined, the prompt remains unchanged, thereby avoiding any breaking changes for users who rely on the default behavior.

I'll proceed with testing the updated code with both the enhanced configuration and the default setup. Please let me know if you have any further suggestions or concerns.

[dhiaaeddine16]

Hi @ccurme, @efriis, @Levara, @tomasonjo could someone please review this PR?

[tomasonjo]

Shouldnt it be just a few line of code that optionally adds props to existing prompt instead of duplicating prompts, and then also chabging position of functions which makes reading the changes harder V ned., 16. feb. 2025, 11:02 je oseba Dhiaa Eddine ***@***.***> napisala:

…

Hi @ccurme <https://github.com/ccurme>, @efriis <https://github.com/efriis>, @Levara <https://github.com/Levara>, @tomasonjo <https://github.com/tomasonjo> could someone please review this PR? — Reply to this email directly, view it on GitHub <#42 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEYGGTMHDWF75SWESUDD32D2QBO43AVCNFSM6AAAAABWO536RKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNRRGM2TGNZVGY> . You are receiving this because you were mentioned.Message ID: ***@***.***> [image: dhiaaeddine16]*dhiaaeddine16* left a comment (langchain-ai/langchain-experimental#42) <#42 (comment)> Hi @ccurme <https://github.com/ccurme>, @efriis <https://github.com/efriis>, @Levara <https://github.com/Levara>, @tomasonjo <https://github.com/tomasonjo> could someone please review this PR? — Reply to this email directly, view it on GitHub <#42 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEYGGTMHDWF75SWESUDD32D2QBO43AVCNFSM6AAAAABWO536RKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNRRGM2TGNZVGY> . You are receiving this because you were mentioned.Message ID: ***@***.***>