Knowledge graph features overview

Foundry IQ (Azure AI Search) provides two graph features for different information needs. Topic graphs broaden evidence across themes, while entity graphs follow connections between specific entities.

Choose a topic graph for diverse queries about themes across a collection. Choose an entity graph for questions that depend on how people, organizations, places, or other entities relate to each other.

Comparison of keyword, vector, topic, and entity retrieval approaches

Important

Knowledge graph retrieval is in public preview and is subject to the Supplemental Terms of Use for Microsoft Azure Previews.

For preview terms specific to this feature, see Preview functionality.

Compare topic graphs and entity graphs

The example queries assume your source content contains the relevant information and that you prepare the graph each feature requires.

Consideration Topic graph Entity graph
What it is Represents topics and their hierarchy. Topic-aware retrieval samples evidence across themes using representative topic vectors. Represents typed entities as nodes and their relationships as edges. Graph queries follow those connections.
What it's good for Diverse queries about themes or patterns across a document collection. Relationship queries, especially questions that require following a chain of connections between entities.
Example queries "What are the main themes across these documents?"
"What topics do the customer feedback documents cover?"
"Which people work for organizations in Seattle?"
"Where are Avery Chen's employers located?"
Ontology requirement No ontology required. An ontology is required to define the entity types and relationships to extract.
Preparation Prepare a topic index with topic identifiers and centroid vectors. A community graph can also represent the topic hierarchy. Define an ontology with the entity types and relationships to extract from source content.
Main limitation Results depend on the topics and centroids represented in the topic index. Results depend on ontology coverage and extraction quality.

Validate with representative questions

Start with one key use case and a small question set before expanding your graph scope. For topic graphs, check whether retrieved evidence covers the relevant themes. For entity graphs, check whether the ontology and extracted relationships support the connections your questions require.

Use these articles to prepare and query each graph: