Translational Generative Retrieval via Potential Query Generation (ICASSP 2025)

This entry summarizes a Chinese forum post cataloging the ICASSP 2025 paper "Translational Generative Retrieval via Potential Query Generation."

Paper metadata

| Field | Content | |-------|---------| | Title | Translational Generative Retrieval via Potential Query Generation, ICASSP 2025 | | Venue | ICASSP 2025 | | Source | IEEE Xplore | | Type | Academic paper |

Key points

• The paper targets information retrieval in the LLM era, where traditional pipelines split retrieval, ranking, and generation into disconnected stages that struggle with natural-language interaction, multi-hop reasoning, and up-to-date knowledge.
• Core scenarios discussed include open-domain retrieval, enterprise knowledge search, conversational search, and recommendation semantics.
• The post frames the paper within the evolution from BM25 → dense dual-tower retrieval → cross-encoders / late interaction → generative retrieval and agentic search, each generation balancing efficiency, effectiveness, and maintainability.
• In the LLM era, new system variables are the inference budget and the action space: whether to retrieve, how many retrieval steps to take, and which tools to call.

Typical method structure (as described in the post)

1. Input and representation — encode queries, documents, and user context into dense/sparse representations or structured prompts. 2. Core modules — retriever, reranker, planner, memory, tool interfaces, arranged sequentially or in parallel. 3. Learning strategies — supervised fine-tuning, contrastive learning, distillation, reinforcement learning (incl. process rewards), synthetic data bootstrapping. 4. Inference strategies — single-pass retrieval, iterative retrieval, parallel sub-queries, early stopping, and budget control.

Evaluation landscape

• Datasets commonly used: MS MARCO, BEIR, Natural Questions, domain corpora, public recommendation sets.
• Metrics: nDCG@10, MRR, Recall@k, Hit@k, human preference, task success rate, latency, and token cost.
• Baselines: BM25, dense retrieval, cross-encoder rerankers, retrieval-free LLMs, commercial search APIs.
• Specific numerical results are not reproduced in this post; consult the original PDF via the IEEE link before citing quantitative claims.

Insights for search / recommendation / personalization

1. Architecture: cascade retrieve–rerank–generate remains mainstream, but agentic paradigms make retrieval count and policy themselves learnable. 2. Data: high-quality instruction data and click/session logs are critical; synthetic data risks leakage and distribution shift. 3. Evaluation: the gap between offline metrics and online satisfaction is widening; LLM-as-judge needs cross-validation with human assessment. 4. Productization: latency, cost, explainability, and safety are hard deployment constraints—do not optimize only for academic benchmarks.

Engineering checklist (from the post's appendix)

| Item | Question | Suggestion | |------|----------|------------| | Data | Does training/indexed data contain PII? Versioning? | Partitioned indexes, anonymization, rollback-able embedding versions | | Latency | What is the p99 budget? How many retrieval steps? | Cascades + early stopping, caching hot queries, async reranking | | Quality | Do offline gains convert to online CTR/satisfaction? | Interleaving experiments, human audits, citation verification | | Security | Does open retrieval introduce poisoning/bias? | Source whitelists, adversarial detection, output filtering | | Cost | Token and GPU cost per query? | Small-model routing, distillation, hybrid sparse+dense |

Related entries cross-referenced in the post

• A Comprehensive Study of Knowledge Editing for Large Language Models (arXiv 2401.01286)
• INTERS: Unlocking the Power of Large Language Models in Search (arXiv 2401.06532)
• Recommendation as Instruction Following (ACM, 10.1145/3708882)
• RouteLLM: Learning to Route LLMs with Preference Data (arXiv 2406.18665)
• Zero-shot Document Retrieval with Hybrid Pseudo-document Retriever (ICASSP)
• Representation Learning with Large Language Models for Recommendation (ACM, 10.1145/3589334.3645458)

Glossary

| Term | Meaning | |------|---------| | IR | Information Retrieval | | RAG | Retrieval-Augmented Generation | | LTR | Learning to Rank | | nDCG | Normalized Discounted Cumulative Gain | | Agentic Search | Search modeled as sequential decision-making and tool use | | Gen-IR | Generative Information Retrieval |

> Note: The original post is a structured catalog entry rather than a full paper translation. Its abstract-quoting section repeats the title without reproducing the actual abstract text, so details above reflect the post's framing and standard methodology descriptions, not verified experimental results from the paper itself.