Step-back prompting

As explained in Essential GraphRAG

Step-back prompting is a prompting technique that reformulates a specific question into a more general question or perspective before retrieval or generation.

Step-back prompting rewrites a detailed user question into a broader, more foundational question before vector retrieval. The broader wording can better match information-rich documents, widen the search, and reduce missed facts caused by query-specific phrasing. section 3.0 section 3.1

Implement the technique as a query-rewriting stage:

• Give an LLM a system instruction to paraphrase the input into a more generic, easier-to-answer question. section 3.1
• Use zero-shot prompting for a direct instruction, or add two to five examples with few-shot prompting to make the transformation more consistent. section 3.1
• Send the original question with the rewriting prompt, then use the generated step-back question for vector retrieval. section 3.1 section 3.0

Career-history retrieval

Rewrite “Which team did Thierry Audel play for from 2007 to 2008?” as “What is the career history of Thierry Audel?” The broader query searches across more comprehensive career information before the system generates the answer. section 3.1

Figure 3.1 Query rewriting by using the step-back technique to increase the vector retrieval accuracy

Shows how a detailed question becomes a broader retrieval query, widening the search before answer generation. section 3.0

How Essential GraphRAG covers it

How deep it goes

Defines the rationale, contrasts zero-shot and few-shot prompting, and provides a prompt and generation function without detailing retrieval evaluation.

Examples

Uses Estella Leopold’s education, Thierry Audel’s career, and prompt examples involving The Police and Jan Sindel.

What it assumes

Assumes vector retrieval and LLM interaction, while teaching the query-rewriting prompt pattern from basic instructions and examples.

What you can do after

Apply it via an LLM prompt and generation function to produce broader questions for subsequent vector retrieval.

How it connects

Connects query rewriting with vector retrieval, zero-shot prompting, few-shot prompting, and hypothetical document retrieval.

liveBookmanning.com

Read Essential GraphRAG on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  AI Agents and ApplicationsStep-back question generation is a technique that reformulates a specific question into a broader question to elicit useful contextual information for retrieval or answering.

Compare how 2 books explain step-back prompting →

Related concepts

• query rewritingkind of · Query rewriting is the transformation of a query into an alternative formulation intended to better support its execution or retrieval while preserving its intended meaning or behavior.
• language modeluses · A language model is a computational model that represents patterns in linguistic tokens to estimate sequence probabilities and predict or generate language from context.
• retrieval-augmented generationused by · Retrieval-augmented generation is a technique that retrieves relevant information from external sources and supplies it as context to a generative model when producing an output.
• promptproduced by · An input, typically containing instructions or context, supplied to an artificial-intelligence model to guide or elicit its output.
• vector databaseuses · A vector database is a data-management system that stores and indexes numerical vector representations and supports retrieval through similarity or distance searches.
• query optimizationkind of · Query optimization is the process of transforming a query or selecting an execution strategy to improve retrieval effectiveness or efficiency while preserving intended results.

This explainer was generated by Manning's knowledge graph from the text of Essential GraphRAG. The section links point to the parts of the book it draws on.