cognee/cognee/tasks/graph/extract_graph_from_data.py
2025-09-17 12:28:44 +02:00

106 lines
3.8 KiB
Python

import asyncio
from typing import Type, List, Optional
from pydantic import BaseModel
from cognee.infrastructure.databases.graph import get_graph_engine
from cognee.tasks.storage.add_data_points import add_data_points
from cognee.modules.ontology.rdf_xml.RDFLibOntologyResolver import RDFLibOntologyResolver
from cognee.modules.chunking.models.DocumentChunk import DocumentChunk
from cognee.modules.graph.utils import (
expand_with_nodes_and_edges,
retrieve_existing_edges,
)
from cognee.shared.data_models import KnowledgeGraph
from cognee.infrastructure.llm.LLMGateway import LLMGateway
from cognee.tasks.graph.exceptions import (
InvalidGraphModelError,
InvalidDataChunksError,
InvalidChunkGraphInputError,
InvalidOntologyAdapterError,
)
async def integrate_chunk_graphs(
data_chunks: list[DocumentChunk],
chunk_graphs: list,
graph_model: Type[BaseModel],
ontology_adapter: RDFLibOntologyResolver,
) -> List[DocumentChunk]:
"""Updates DocumentChunk objects, integrates data points and edges into databases."""
if not isinstance(data_chunks, list) or not isinstance(chunk_graphs, list):
raise InvalidChunkGraphInputError("data_chunks and chunk_graphs must be lists.")
if len(data_chunks) != len(chunk_graphs):
raise InvalidChunkGraphInputError(
f"length mismatch: {len(data_chunks)} chunks vs {len(chunk_graphs)} graphs."
)
if not isinstance(graph_model, type) or not issubclass(graph_model, BaseModel):
raise InvalidGraphModelError(graph_model)
if ontology_adapter is None or not hasattr(ontology_adapter, "get_subgraph"):
raise InvalidOntologyAdapterError(
type(ontology_adapter).__name__ if ontology_adapter else "None"
)
graph_engine = await get_graph_engine()
if graph_model is not KnowledgeGraph:
for chunk_index, chunk_graph in enumerate(chunk_graphs):
data_chunks[chunk_index].contains = chunk_graph
return data_chunks
existing_edges_map = await retrieve_existing_edges(
data_chunks,
chunk_graphs,
)
graph_nodes, graph_edges = expand_with_nodes_and_edges(
data_chunks, chunk_graphs, ontology_adapter, existing_edges_map
)
if len(graph_nodes) > 0:
await add_data_points(graph_nodes)
if len(graph_edges) > 0:
await graph_engine.add_edges(graph_edges)
return data_chunks
async def extract_graph_from_data(
data_chunks: List[DocumentChunk],
graph_model: Type[BaseModel],
ontology_adapter: RDFLibOntologyResolver = None,
custom_prompt: Optional[str] = None,
) -> List[DocumentChunk]:
"""
Extracts and integrates a knowledge graph from the text content of document chunks using a specified graph model.
"""
if not isinstance(data_chunks, list) or not data_chunks:
raise InvalidDataChunksError("must be a non-empty list of DocumentChunk.")
if not all(hasattr(c, "text") for c in data_chunks):
raise InvalidDataChunksError("each chunk must have a 'text' attribute")
if not isinstance(graph_model, type) or not issubclass(graph_model, BaseModel):
raise InvalidGraphModelError(graph_model)
chunk_graphs = await asyncio.gather(
*[
LLMGateway.extract_content_graph(chunk.text, graph_model, custom_prompt=custom_prompt)
for chunk in data_chunks
]
)
# Note: Filter edges with missing source or target nodes
if graph_model == KnowledgeGraph:
for graph in chunk_graphs:
valid_node_ids = {node.id for node in graph.nodes}
graph.edges = [
edge
for edge in graph.edges
if edge.source_node_id in valid_node_ids and edge.target_node_id in valid_node_ids
]
return await integrate_chunk_graphs(
data_chunks, chunk_graphs, graph_model, ontology_adapter
)