Assume a dummy document for graph hits [chapter] deterministic

merged[name] = {"page_content": f"Graph node: {name}"} return list(merged.values())[:top_k * 2]

`

The hybrid retriever returns both text chunks and graph node identifiers. You can then pass these to the LLM with a prompt that explicitly references the graph context.

Using Graph Context in the Prompt To keep the LLM grounded, include the retrieved graph paths in the prompt. For instance:

`

You are an that answers questions using the provided context. Graph context: {graph_context} Text context: {text_context}

`

The LLM will then be able to reason over the graph edges, citing specific entities like BlogGenerator Wiki Page or Cola. Because the graph is local, you can also enforce that the model only uses entities present in the graph, reducing the risk of fabricating relationships.

Evaluating Graph‑Augmented RAG Metrics such as Faithfulness, Answer Relevance, and Context Recall can be computed on a held‑out set of queries. A common evaluation involves: 1. Generating answers with the hybrid retriever. 2. Using an LLM judge to compare the answer against the ground truth and the retrieved context. 3. Measuring the proportion of answers that correctly reference graph nodes. If the graph improves answer accuracy, you have empirical evidence that the knowledge graph is adding value.

Visualizing and Querying Knowledge Graphs

Visualizing the Graph Visualization helps you understand the structure, spot isolated clusters, and communicate the graph to stakeholders. Tools like pyvis and networkx work well for small graphs; for larger datasets, consider neo4j Bloom or Gephi. Below is a minimal example using pyvis to create an interactive HTML visualization from a networkx graph.

```python import networkx as nx from pyvis.network import Network def network_to_pyvis(G: nx.Graph): net = Network(notebook=True, directed=False) net.from_nx(G) return net