# Evolving Semantic Indexes for Generative Retrieval

DevFeed: [Evolving Semantic Indexes for Generative Retrieval](<https://devfeed.tech/articles/evolving-semantic-indexes-for-generative-retrieval-94439.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.28568>)

Author: Jinming Nian, Jyun-Yu Jiang, Henry Tsai, Yi Fang, Zhirong Liang

Published: 2026-10-21T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [AI-driven information retrieval](<https://devfeed.tech/topics/ai-driven-information-retrieval.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [api](<https://devfeed.tech/tags/api.md>), [application](<https://devfeed.tech/tags/application.md>), [approach](<https://devfeed.tech/tags/approach.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [autoregressive](<https://devfeed.tech/tags/autoregressive.md>), [but](<https://devfeed.tech/tags/but.md>), [commerce](<https://devfeed.tech/tags/commerce.md>), [component](<https://devfeed.tech/tags/component.md>), [components](<https://devfeed.tech/tags/components.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [decoder](<https://devfeed.tech/tags/decoder.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [design](<https://devfeed.tech/tags/design.md>), [e-commerce](<https://devfeed.tech/tags/e-commerce.md>), [evolution](<https://devfeed.tech/tags/evolution.md>), [generative](<https://devfeed.tech/tags/generative.md>), [indexes](<https://devfeed.tech/tags/indexes.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [semantic](<https://devfeed.tech/tags/semantic.md>)

## AI overview

Generative retrieval relies on discrete semantic identifiers whose interdependent design is difficult to optimize with gradients. The paper uses evolutionary program search guided by a locally served open-weight 31B model, with a 1M-parameter proxy decoder to evaluate candidates without repeatedly training the 1B target generator. Across 21 SID programs on NQ320K, proxy and target Recall@100 rankings correlate strongly (Spearman ρ=0.90). Evolving SID construction and collision handling raises NQ320K Recall@1 from 55.8% to 59.3% and Recall@100 from 85.8% to 90.6%; improvements are also reported on MSMARCO300K and a multilingual e-commerce dataset under a fixed training recipe. Transfer depends on the search space: proxy-guided improvements were reliable for the semantic index, but not for the generator training objective or the full pipeline, so proxy reliability must be checked for each chosen space.

## Source excerpt

arXiv:2610.28568v1 Announce Type: cross Abstract: Generative retrieval (GR) casts search and recommendation as autoregressive generation of discrete semantic identifiers (SIDs). SID design largely determines GR quality, but its discrete, interdependent components (SID construction, collision handling, decoding, and training) cannot be optimized end-to-end with gradients and remain largely hand-designed. We introduce a framework that improves these components through LLM-driven evolutionary program search, to our knowledge the first application of this approach to GR. A locally served, open-weight 31B model mutates executable GR components, so no proprietary LLM API is used, and a 1M-parameter proxy decoder provides fast feedback in place of repeatedly training the 1B target generator. Across 21 SID programs on NQ320K, proxy and target Recall@100 show strong rank correlation (Spearman $\rho = 0.90$). Guided by this proxy, evolution improves the seed GR on NQ320K, MSMARCO300K, and a multilingual e-commerce dataset under a fixed training recipe. On NQ320K, jointly evolving SID construction and collision handling improves Recall@1 from 55.8% to 59.3% and Recall@100 from 85.8% to 90.6%. A component study shows that proxy reliability depends on what is evolved: improvements transfer reliably for the semantic index but not for the generator's training objective or the full pipeline. LLM-driven evolutionary search is thus a practical tool for GR design, provided proxy reliability is established for the chosen search space.