Xetrieval: Mechanistically Explaining Dense Retrieval

Zhixin Cai, Jun Bai, Yang Liu, Jiaqi Li, Yichi Zhang, Taichuan Li, Zhuofan Chen, Zixia Jia, Zilong Zheng, Wenge Rong

arXiv:2605.29507 · 2026-05-29 공개 · arXiv · PDF

chain-of-thought embedding-space feature-decomposition retrieval-explanation reasoning-internalizer interpretable-features retrieval-intervention natural-language-description

Abstract

Explaining why dense retrievers assign high relevance scores remains challenging because retrieval decisions are made through opaque high-dimensional embeddings. Existing explanations often focus on surface signals, such as lexical matches, token alignments, or post-hoc textual rationales, and thus provide limited insight into the latent factors that shape dense retrieval behavior at the embedding level. We propose Xetrieval, an embedding-level mechanistic framework for explaining dense retrieval. Xetrieval first introduces a lightweight reasoning internalizer that approximates Chain-of-Thought reasoning directly in the embedding space with a single forward pass, enriching sentence embeddings with reasoning-oriented information while avoiding expensive autoregressive generation. It then decomposes these reasoning-enhanced embeddings into sparse, human-interpretable features, each associated with a coherent natural language description. By aggregating sparse feature overlaps across multiple document-side views, Xetrieval provides feature-level explanations of individual retrieval decisions. Experiments on diverse retrievers and benchmarks show that Xetrieval uncovers coherent interpretable features, yields stronger pair-level intervention effects, and supports task-level feature steering. The project page and source code are available at https://hihiczx.github.io/Xetrieval .

한국어 요약

📋 한 줄 요약

**[Dense Retrieval / Mechanistic Explanation]** Xetrieval이 reasoning internalizer로 CoT를 single forward pass로 embedding에 주입 후 sparse interpretable feature로 분해 — 검색 결정의 feature-level 설명·intervention·task-level steering 가능.

🎯 핵심 기여도

💡 핵심 아이디어

Dense retrieval 설명을 lexical match 수준이 아닌 embedding space의 mechanistic feature로 끌어올리려면 CoT-style reasoning을 embedding에 single-pass로 internalize하고, 이를 sparse·자연어로 설명되는 feature로 분해해 multi-view 집계해야 한다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: Dense retrieval의 embedding-level mechanistic 설명의 첫 통합 프레임워크, CoT를 single forward pass로 근사하는 효율 설계, feature-level 설명·intervention·steering까지 확장된 실용성. **한계**: Reasoning internalizer의 학습 데이터·품질 의존, sparse feature decomposition의 sparsity·해석성 trade-off, 매우 도메인 specific retrieval에서의 일반화는 후속.

🚀 실용적 활용