GraphRAG

Question Generation ❔

Entity-based Question Generation

The question generation method combines structured data from the knowledge graph with unstructured data from the input documents to generate candidate questions related to specific entities.

Methodology

Given a list of prior user questions, the question generation method uses the same context-building approach employed in local search to extract and prioritize relevant structured and unstructured data, including entities, relationships, covariates, community reports and raw text chunks. These data records are then fitted into a single LLM prompt to generate candidate follow-up questions that represent the most important or urgent information content or themes in the data.

Configuration

Below are the key parameters of the Question Generation class:

How to Use

An example of the question generation function can be found in the following notebook.