Xetrieval: Mechanistically Explaining Dense Retrieval

2026-05-28

AI Summary

Dense retrieval models suffer from limited interpretability due to their reliance on high-dimensional embeddings, obscuring the underlying decision mechanisms. This work proposes Xetrieval, a novel framework that provides the first mechanistic explanation of dense retrieval at the embedding level. Xetrieval employs a lightweight reasoning internalizer to approximate chain-of-thought reasoning within the embedding space and decomposes enhanced embeddings into sparse, interpretable semantic features. Requiring only a single forward pass, the method leverages multi-view feature overlap aggregation to enable sample-level interventions and task-oriented feature manipulation. Experiments across multiple retrievers and benchmarks demonstrate that Xetrieval consistently uncovers coherent and interpretable features, significantly improving intervention efficacy and model controllability.

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 \textit{Xetrieval}, an embedding-level mechanistic framework for explaining dense retrieval. \textit{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, \textit{Xetrieval} provides feature-level explanations of individual retrieval decisions. Experiments on diverse retrievers and benchmarks show that \textit{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 .

Problem

• dense retrieval
• interpretability
• embedding-level explanation
• relevance scoring
• mechanistic explanation

Innovation

• dense retrieval
• mechanistic explanation
• embedding-level interpretation
• Chain-of-Thought reasoning
• sparse feature decomposition