Interpret and Control Dense Retrieval with Sparse Latent Features

2024-10-17

arXiv.org

AI Summary

To address the poor interpretability and limited controllability of dense retrieval models, this paper proposes a latent-space modeling approach based on sparse autoencoders (SAEs), coupled with a retrieval-oriented contrastive loss that learns semantically faithful and task-effective sparse latent features. Our method enables, for the first time, fine-grained, controllable editing of dense retrieval behavior—e.g., steering results along dimensions such as topic, recency, or sentiment—while preserving high interpretability and strong controllability. On benchmarks including MSMARCO, reconstructed vectors retain over 99% of the original dense representation’s retrieval accuracy. Empirical results demonstrate that the learned sparse features exhibit clear semantic correspondence with human-understandable concepts and support precise, targeted interventions in the latent space. This work bridges the gap between dense retrieval performance and transparent, user-controllable behavior.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Dense embeddings deliver strong retrieval performance but often lack interpretability and controllability. This paper introduces a novel approach using sparse autoencoders (SAE) to interpret and control dense embeddings via the learned latent sparse features. Our key contribution is the development of a retrieval-oriented contrastive loss, which ensures the sparse latent features remain effective for retrieval tasks and thus meaningful to interpret. Experimental results demonstrate that both the learned latent sparse features and their reconstructed embeddings retain nearly the same retrieval accuracy as the original dense vectors, affirming their faithfulness. Our further examination of the sparse latent space reveals interesting features underlying the dense embeddings and we can control the retrieval behaviors via manipulating the latent sparse features, for example, prioritizing documents from specific perspectives in the retrieval results.

Problem

• Interpret dense embeddings' latent features
• Control retrieval using sparse autoencoders
• Maintain retrieval accuracy with interpretability

Innovation

• Sparse autoencoders interpret dense embeddings
• Contrastive loss ensures retrieval effectiveness
• Manipulate latent features for controlled retrieval