SHIFT: Self-reconstruction Harnesses Implicit Fine-grained Thinking for Retrieval

2026-07-23

AI Summary

This work addresses the performance limitations of existing large language model (LLM)-based retrievers in reasoning-intensive tasks, which often stem from misalignment between retrieval and generation objectives. To bridge this gap, the authors propose a novel framework that efficiently transforms LLMs into reasoning-aware retrievers through residual projection and task-oriented bidirectional attention aggregation. Additionally, they introduce a self-reconstruction mechanism grounded in fine-grained next-token prediction to explicitly align contrastive learning with implicit reasoning objectives. By integrating implicit fine-grained reasoning with self-reconstruction training, the approach effectively narrows the semantic gap between retrieval and generation. Experimental results demonstrate that the proposed method significantly outperforms state-of-the-art retrievers across multiple reasoning-intensive retrieval benchmarks, confirming its effectiveness and robustness.

Abstract

LLM-based retrievers have become a fundamental component of modern information retrieval systems. The paradigm of "rewrite-then-retriev" introduces explicit reasoning before retrieval. In addition, implicit-reasoning retrievers such as GIRCSE and LaSER improve efficiency by replacing explicit reasoning with soft tokens. Although these methods demonstrated competitive performance on reasoning-intensive retrieval benchmarks, they struggle to address the mismatch between the objectives of retrieval and generation. In this work, we propose SHIFT ($\textbf{S}$elf-reconstruction $\textbf{H}$arnesses $\textbf{I}$mplicit $\textbf{F}$ine-grained $\textbf{T}$hinking for Retrieval), a retrieval training framework based on LLMs. Firstly, we transfer LLMs into reasoning-efficient retrievers with residual projection and task-oriented bidirectional attention aggregation in the latent space. Secondly, we alleviate the mismatch between contrastive learning and implicit reasoning using fine-grained next-token-prediction-based reconstruction. Extensive experiments on reasoning-intensive retrieval benchmarks show that SHIFT consistently outperforms other widely used retrievers. We also carried out a detailed analysis to illustrate how our method works.

Problem

• retrieval
• reasoning
• mismatch
• LLM
• implicit reasoning

Innovation

• implicit reasoning
• self-reconstruction
• residual projection
• bidirectional attention aggregation
• LLM-based retriever