Building Knowledge Graphs from Documents with LLM using CocoIndex

Key Concepts:

• CocoIndex: A tool for building and maintaining knowledge graphs with continuous source updates.
• LLM (Large Language Model): Used to extract relationships between concepts in documents.
• Knowledge Graph: A structured representation of knowledge, showing relationships between entities.

Types of Relationships Extracted:

1. Subject-Object Relationships: E.g., "CocoIndex supports Incremental Processing."
2. Entity Mentions: E.g., "core/basics.mdx" mentions CocoIndex and Incremental Processing.

Prerequisites:

• PostgreSQL (for CocoIndex's incremental processing)
• Neo4j (graph database)
• OpenAI API key (or Ollama for local LLM models)

Data Flow:

1. Add Documents as Source: CocoIndex documentation markdown files (.md, .mdx).
2. Add Data Collectors:
   • document_node: Collects documents (e.g., core/basics.mdx).
   • entity_relationship: Collects relationships (e.g., "CocoIndex supports Incremental Processing").
   • entity_mention: Collects mentions of entities in a document (e.g., core/basics.mdx mentions CocoIndex).
3. Process Each Document and Extract Summary: Use cocoindex.functions.ExtractByLlm with gpt-4o to get document title and summary.
4. Extract Relationships: Use cocoindex.functions.ExtractByLlm to extract relationships (subject, predicate, object) from the document.
5. Collect Relationships: Collect relationships between subjects and objects and mentions of entities in the document.
6. Build Knowledge Graph: Export nodes and relationships to Neo4j.

Building the Knowledge Graph in Neo4j:

• Nodes: Represent entities (e.g., Document, Entity). Each node needs a label and a primary key field.
• Relationships: Connect nodes.

Exporting to Neo4j:

1. Configure Neo4j Connection: Specify URI, user, and password.
2. Export Document Nodes: Export from the document_node collector, using filename as the primary key.
3. Export RELATIONSHIP and Entity Nodes: Declare Entity nodes and then export relationships from the entity_relationship collector.
4. Export entity_mention: Creates Document nodes and Entity nodes and connects them with MENTION relationships.

Code Snippets:

• Adding a source:

  flow_builder.add_source(cocoindex.sources.LocalFile(path="../../docs/docs/core", included_patterns=["*.md", "*.mdx"]))
  

• Extracting summary:

  doc["summary"] = doc["content"].transform(cocoindex.functions.ExtractByLlm(llm_spec=cocoindex.LlmSpec(api_type=cocoindex.LlmApiType.OPENAI, model="gpt-4o"), output_type=DocumentSummary, instruction="Please summarize the content of the document."))
  

• Extracting relationships:

  doc["relationships"] = doc["content"].transform(cocoindex.functions.ExtractByLlm(llm_spec=cocoindex.LlmSpec(api_type=cocoindex.LlmApiType.OPENAI, model="gpt-4o"), output_type=list[Relationship], instruction=("Please extract relationships from CocoIndex documents. " "Focus on concepts and ignore examples and code. ")))
  

• Exporting Document nodes to Neo4j:

  document_node.export("document_node", cocoindex.storages.Neo4j(connection=conn_spec, mapping=cocoindex.storages.Nodes(label="Document")), primary_key_fields=["filename"])
  

• Exporting relationships and entities to Neo4j:

  entity_relationship.export("entity_relationship", cocoindex.storages.Neo4j(connection=conn_spec, mapping=cocoindex.storages.Relationships(rel_type="RELATIONSHIP", source=cocoindex.storages.NodeFromFields(label="Entity", fields=[cocoindex.storages.TargetFieldMapping(source="subject", target="value")]), target=cocoindex.storages.NodeFromFields(label="Entity", fields=[cocoindex.storages.TargetFieldMapping(source="object", target="value")])), primary_key_fields=["id"])
  

• Exporting the entity_mention to Neo4j:

  entity_mention.export("entity_mention", cocoindex.storages.Neo4j(connection=conn_spec, mapping=cocoindex.storages.Relationships(rel_type="MENTION", source=cocoindex.storages.NodesFromFields(label="Document", fields=[cocoindex.storages.TargetFieldMapping("filename")]), target=cocoindex.storages.NodesFromFields(label="Entity", fields=[cocoindex.storages.TargetFieldMapping(source="entity", target="value")]))), primary_key_fields=["id"])
  

Running the Index:

1. Install dependencies: pip install -e .

2. Setup and update the index:

   python main.py cocoindex setup
   python main.py cocoindex update
   

Querying the Knowledge Graph:

You can explore the knowledge graph in Neo4j Browser using Cypher queries.

MATCH p=()-->() RETURN p
``