feat(llm): property embedding

[MrJs133]

link to #228, V1 version

update, get and delete property embeddings

• Added property embedding updates to update_vid_embedding (embedding will only be performed for indexed properties).
  • Adopt incremental updates (support adding and deleting prop_value, as well as adding and deleting prop_key). If the number of properties to be updated exceeds 100,000, a warning will be issued, and the update will be rejected.
• Added property embedding removes to clean_all_graph_index

keywords match

Match properties according to keywords (exact + fuzzy)

text2gremlin

Add properties in the prompt

subgraph_query

Give priority to executing the vid subgraph query. If it fails and properties have been matched based on keywords previously, then execute the property subgraph query.

[Copilot]

Pull Request Overview

This PR introduces property embedding support and extends existing vector indexing and query functionality. Key changes include:

• Adding property vector indexing and cleaning in the vector and graph index utilities.
• Updating embedding retrieval and query generation to support property keywords and property embeddings.
• Integrating property-related logic into semantic queries, index building, API endpoints, and prompt configuration.

Reviewed Changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
hugegraph-llm/src/hugegraph_llm/utils/vector_index_utils.py	Added property vector index loading and updated index info output.
hugegraph-llm/src/hugegraph_llm/utils/graph_index_utils.py	Incorporated cleaning and update logging for property vector indexes.
hugegraph-llm/src/hugegraph_llm/operators/llm_op/gremlin_generate.py	Passed an additional 'properties' parameter to LLM generation.
hugegraph-llm/src/hugegraph_llm/operators/index_op/semantic_id_query.py	Integrated exact and fuzzy matching for property keys and values.
hugegraph-llm/src/hugegraph_llm/operators/index_op/build_semantic_index.py	Added diffing and update logic for property embeddings with limit checks.
hugegraph-llm/src/hugegraph_llm/operators/hugegraph_op/graph_rag_query.py	Updated gremlin queries to use property name/value filtering in fallback scenarios.
hugegraph-llm/src/hugegraph_llm/operators/hugegraph_op/fetch_graph_data.py	Extended graph summary with extracted index label properties.
hugegraph-llm/src/hugegraph_llm/operators/gremlin_generate_task.py	Passed property data to the gremlin generate operator.
hugegraph-llm/src/hugegraph_llm/demo/rag_demo/app.py	Registered and imported the new vector API endpoint.
hugegraph-llm/src/hugegraph_llm/config/prompt_config.py	Updated prompt content to include property references for query generation.
hugegraph-llm/src/hugegraph_llm/api/vector_api.py	Introduced new API endpoint for updating vector embeddings with rate limits.

Comments suppressed due to low confidence (1)

hugegraph-llm/src/hugegraph_llm/operators/hugegraph_op/graph_rag_query.py:349

• The variable 'use_id_to_match' is used without a visible definition in this context. Please ensure it is properly declared and initialized before use.

node_str = f"{item['id']}{{{props_str}}}" if use_id_to_match else f"{item['props']}{{{props_str}}}"

[Copilot]

[nitpick] Consider refactoring the hardcoded property update limit (100000) into a configurable parameter to improve maintainability and flexibility.

[Copilot]

Ensure that 'get_texts_embeddings' returns a non-empty list before accessing its first element to avoid potential IndexError.

Suggested change

	keyword_vector = self.embedding.get_texts_embeddings([keyword])
	keyword_vector = self.embedding.get_texts_embeddings([keyword])
	if not keyword_vector: # Ensure the list is non-empty
	log.warning("No embeddings found for keyword: %s", keyword)
	continue