How Does Generative Retrieval Scale to Millions of Passages?
Topics: Indexing, LLMO / GEO, Passage based retrieval, Query Fan Out, Ranking, Retrieval Augmented Generation (RAG)
This Google paper investigates how generative retrieval techniques perform when scaled to millions of passages. Unlike traditional retrieval systems that rely on external indices, generative retrieval reframes retrieval as a sequence-to-sequence problem, mapping queries to document identifiers. The study empirically evaluates generative retrieval methods on various corpus sizes, including the full MS MARCO dataset with 8.8M passages. Key findings highlight the importance of synthetic query generation (query fan out) , the limitations of naive model scaling, and the ineffectiveness of existing architecture modifications when considering compute cost. The results suggest that while generative retrieval is competitive with dual encoders on small datasets, scaling to millions of passages remains an open challenge.
Your advantages:
+ Get access to the full exclusive paid articles in the blog.
+ Full analysis of hundreds of well researched active Microsoft and Google patents and research paper.
+ Save a lot of time and get insights in just a few minutes, without having to spend hours analyzing the documents.
+ Get quick exclusive insights about how search engines and Google could work with easy to understand summaries and analysis.
+ All patents classified by topic for targeted research.
+ New patent summaries and analysis every week. Weekly notification via E-Mail
+ Use all 4 AI Research Tools to gain insights in seoncds from all documents in the taining databases, the Google Leak Analyzer, Patent & Paper Analyzer, Semantic SEO Research Agent, LLMO / GEO Assistant
+ Gain fundamental insights for your SEO work and become a real thought leader.
Get access to the SEO Research Suite and become a SEO thought leader now!
