liveBook · Manning

Asymmetric search

As explained in Retrieval Augmented Generation, The Foundational Ideas

Asymmetric search retrieves documents or passages for queries whose length, structure, and communicative role differ substantially from the targets.

Asymmetric search retrieves a long, specific passage for a short, interrogative, and often vague query. The query and document may answer one another despite sharing little surface form, unlike symmetric search, which compares texts of similar length and structure. section 3.1

The central design contrast is:

  • Sparse representations support efficient keyword matching but miss semantic meaning; dense representations encode semantic context across continuous dimensions. section 3.1
  • SBERT targets symmetric comparisons such as duplicate questions or paraphrases, whereas DPR uses separate BERT encoders for questions and passages. section 3.1
  • DPR learns the asymmetric space from question, correct-passage, and incorrect-passage triplets for open-domain question answering. section 3.1
Table 3.1 Symmetric search vs asymmetric search
  Feature  
  Symmetric Search  
  Asymmetric Search  
 Goal 
 Find similar text (twins) 
 Find answers to questions 
 Input 
 Similar length/structure 
 Short query vs. Long doc 
 Model 
 SBERT 
 DPR 
 Use Case 
 Paraphrase detection 
 Open Domain QA 

Compare symmetric and asymmetric search by goal, input shape, model, and use case. section 3.1

How Retrieval Augmented Generation, The Foundational Ideas covers it

How deep it goes
Explains representation differences, model specialization, and triplet training without detailing retrieval implementation.
Examples
Uses question-to-passage retrieval, duplicate-question detection, paraphrase comparison, and sparse-versus-dense vector listings.
What you can do after
Evaluate whether symmetric SBERT or asymmetric DPR better matches a question-to-passage retrieval problem.
How it connects
Connects asymmetric search to dense vectors, sparse vectors, BM25, SBERT, DPR, and open-domain question answering.

This explainer was generated by Manning's knowledge graph from the text of Retrieval Augmented Generation, The Foundational Ideas. The section links point to the parts of the book it draws on.