Efficient Temporal-aware Matryoshka Adaptation for Temporal Information Retrieval

TL;DR

This paper introduces TMRL, a novel method that enhances temporal relevance in retrieval systems for Temporal RAG by integrating temporal-aware Matryoshka embeddings, improving efficiency and performance.

Contribution

It presents a new temporal-aware embedding technique that adapts existing models for better temporal retrieval in RAG systems, balancing accuracy and efficiency.

Findings

TMRL improves temporal retrieval accuracy over non-temporal methods.

TMRL achieves competitive performance with prior temporal methods.

The approach enables flexible trade-offs between accuracy and efficiency.

Abstract

Retrievers are a key bottleneck in Temporal Retrieval-Augmented Generation (RAG) systems: failing to retrieve temporally relevant context can degrade downstream generation, regardless of LLM reasoning. We propose Temporal-aware Matryoshka Representation Learning (TMRL), an efficient method that equips retrievers with temporal-aware Matryoshka embeddings. TMRL leverages the nested structure of Matryoshka embeddings to introduce a temporal subspace, enhancing temporal encoding while preserving general semantic representations. Experiments show that TMRL efficiently adapts diverse text embedding models, achieving competitive temporal retrieval and temporal RAG performance compared to prior Matryoshka-based non-temporal methods and prior temporal methods, while enabling flexible accuracy-efficiency trade-offs.