Large Scale Retrieval for the LinkedIn Feed using Causal Language Models (arXiv 2510.14223)

Forum topic · 小凯 · 2026-07-05

Summary

This arXiv paper (2510.14223, October 2025), authored by a 23-person team at LinkedIn including Sudarshan Srinivasa Ramanujam, Antonio Alonso, Saurabh Kataria, and colleagues, describes how LinkedIn built large-scale retrieval for its Feed system using causal language models. The work sits at the intersection of information retrieval, recommendation, and LLM-based systems, addressing how to redeploy retrieval, ranking, and generation responsibilities in the era of large language models. The forum post surrounding the paper provides contextual analysis rather than the full abstract: it outlines the typical pipeline of such systems (dense/sparse representation learning, retriever, reranker, planner, generator, and feedback mechanisms), common training strategies (supervised fine-tuning, contrastive learning, distillation, reinforcement learning), and inference tactics such as iterative retrieval and budget control. It also discusses engineering constraints like p99 latency budgets, embedding versioning, safety against corpus poisoning, and per-query token cost. Quantitative results should be verified against the original PDF. The paper is relevant to readers following retrieval-augmented generation, generative retrieval, and industrial-scale personalization systems.

Large Scale Retrieval for the LinkedIn Feed using Causal Language Models

Source: https://arxiv.org/abs/2510.14223 (arXiv, October 2025)

Authors: Sudarshan Srinivasa Ramanujam, Antonio Alonso, Saurabh Kataria, Siddharth Dangi, Akhilesh Gupta, Birjodh Singh Tiwana, et al. (23 authors total, LinkedIn)

Category: Retrieval

Overview

This paper presents LinkedIn's approach to large-scale retrieval for the Feed using causal language models. The work addresses a central question for modern search, recommendation, and personalization systems: how to redistribute the responsibilities of retrieval, ranking, generation, and tool calling in the LLM era.

Note: this forum entry is based on the paper's public metadata and a contextual analysis; the original abstract text was not reproduced in the source post. Consult the PDF for exact methods and quantitative results.

Context and Motivation

Traditional pipelines treat retrieval, ranking, and generation as separate stages, which struggles to meet LLM-era demands for natural language interaction, multi-hop reasoning, and up-to-date knowledge. Core scenarios include open-domain retrieval, enterprise knowledge search, conversational search, and semantic understanding in recommendation systems.

Typical Architecture (as analyzed in the post)

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

Engineering Checklist

| Concern | Question | Suggestion | |---------|----------|------------| | Data | PII in training/index? Version control? | Partitioned indexes, sanitization, rollback-able embedding versions | | Latency | p99 budget? Retrieval steps? | Cascades + early stop, cache popular queries, async reranking | | Quality | Do offline gains transfer to online CTR/satisfaction? | Interleaving experiments, human audits, citation checks | | Safety | Poisoning/bias from open retrieval? | Source whitelists, adversarial detection, output filtering | | Cost | Token and GPU cost per query? | Small-model routing, distillation, hybrid sparse+dense |

Takeaways for Search / Rec / Personalization

Architecture: cascaded retrieve–rerank–generate remains mainstream, but agentic paradigms make retrieval count and policy learnable.
Data: high-quality instruction data and click/session logs matter as much as model size; synthetic data risks leakage and distribution shift.
Evaluation: the gap between offline metrics and online satisfaction is widening; LLM-as-judge needs cross-validation with human evaluation.
Production: latency, cost, interpretability, and safety are hard constraints — don't optimize only academic benchmarks.

Limitations and Open Problems

Potential limitations include experiment scale constrained by GPU budgets, benchmark–production distribution mismatch, English-centric data, and safety risks of agentic systems on the open web. Future directions: better test-time compute allocation, deeper integration with knowledge graphs/structured databases, and causal/fairness constraints for recommenders.

Related Entries

CAME: Competitively Learning a Mixture-of-Experts Model for First-stage Retrieval
CoEvo: Coevolution of LLM and Retrieval Model for Domain-Specific Information Retrieval
DRAMA: Diverse Augmentation from Large Language Models to Smaller Dense Retrievers (arXiv 2502.18460)
ExpandR: Teaching Dense Retrievers Beyond Queries with LLM Guidance
Fine-Tuning LLaMA for Multi-Stage Text Retrieval (arXiv 2310.08319)
How Does Generative Retrieval Scale to Millions of Passages? (Google, arXiv 2305.11841)