Overview

​

Knowledge Graph Provider

R2R supports knowledge graph functionality to enhance document understanding and retrieval. By default, R2R uses Neo4j as the knowledge graph provider. We are actively working to integrate with Memgraph. You can find out more about creating knowledge graphs in the GraphRAG Cookbook.

To configure the knowledge graph settings for your project:

1. Edit the kg section in your r2r.toml file:

[kg]
provider = "neo4j"
batch_size = 256
kg_extraction_prompt = "graphrag_triplet_extraction_zero_shot"

  [kg.kg_creation_settings]
    entity_types = [] # if empty, all entities are extracted
    relation_types = [] # if empty, all relations are extracted
    generation_config = { model = "gpt-4o-mini" }
    max_knowledge_triples = 100 # max number of triples to extract for each document chunk
    fragment_merge_count = 4 # number of fragments to merge into a single extraction

  [kg.kg_enrichment_settings]
    max_description_input_length = 65536 # increase if you want more comprehensive descriptions
    max_summary_input_length = 65536
    generation_config = { model = "gpt-4o-mini" } # and other generation params below
    leiden_params = { max_levels = 10 } # more params in https://neo4j.com/docs/graph-data-science/current/algorithms/leiden/

  [kg.kg_search_config]
    model = "gpt-4o-mini"

Let’s break down the knowledge graph configuration options:

• provider: Specifies the knowledge graph provider. Currently, “neo4j” is supported.
• batch_size: Determines the number of entities or relationships to process in a single batch during import operations.
• kg_extraction_prompt: Specifies the prompt template to use for extracting knowledge graph information from text.
• kg_creation_settings: Configuration for the model used in knowledge graph creation.
  • max_knowledge_triples: The maximum number of knowledge triples to extract for each document chunk.
  • fragment_merge_count: The number of fragments to merge into a single extraction.
  • generation_config: Configuration for the model used in knowledge graph creation.
• kg_enrichment_settings: Similar configuration for the model used in knowledge graph enrichment.
  • generation_config: Configuration for the model used in knowledge graph enrichment.
  • leiden_params: Parameters for the Leiden algorithm.
• kg_search_config: Similar configuration for the model used in knowledge graph search operations.

​

Neo4j Configuration

When using Neo4j as the knowledge graph provider, you need to set up the following environment variables or provide them in the r2r.toml file. To set them as environment variables:

export NEO4J_USER=your_neo4j_username
export NEO4J_PASSWORD=your_neo4j_password
export NEO4J_URL=bolt://your_neo4j_host:7687
export NEO4J_DATABASE=neo4j

And to set them directly in your config:

[kg]
provider = "neo4j"
user = "your_neo4j_username"
password = "your_neo4j_password"
url = "bolt://your_neo4j_host:7687"
database = "neo4j"

​

Knowledge Graph Operations

The Neo4jKGProvider supports various operations:

1. Entity Management: Add, update, and retrieve entities in the knowledge graph.
2. Relationship Management: Create and query relationships between entities.
3. Batch Import: Efficiently import large amounts of data using batched operations.
4. Vector Search: Perform similarity searches on entity embeddings.
5. Community Detection: Identify and manage communities within the graph.

​

Customization

You can customize the knowledge graph extraction and search processes by modifying the kg_extraction_prompt and adjusting the model configurations in kg_extraction_config and kg_search_config. Moreover, you can customize the LLM models used in various parts of the knowledge graph creation process. All of these options can be selected at runtime, with the only exception being the specified database provider. For more details, refer to the knowledge graph settings in the search API.

By leveraging the knowledge graph capabilities, you can enhance R2R’s understanding of document relationships and improve the quality of search and retrieval operations.

​

Next Steps

For more detailed information on configuring specific components of the ingestion pipeline, please refer to the following pages:

• Ingestion Configuration
• Enrichment Configuration
• Retrieval Configuration