Efficient Temporal-aware Matryoshka Adaptation for Temporal Information Retrieval

2026-01-09

arXiv.org

AI Summary

This work addresses the challenge in Temporal Retrieval-Augmented Generation (Temporal RAG) systems, where retrieval components often fail to capture temporal information, thereby degrading generation quality. To this end, the authors propose Temporal-aware Matryoshka Representation Learning (TMRL), which for the first time integrates the nested structure of Matryoshka embeddings with temporal modeling to construct temporal subspaces that enhance temporal encoding while preserving general semantic representations. TMRL supports adaptive fine-tuning across various text embedding models, enabling a flexible trade-off between accuracy and efficiency. Experimental results demonstrate that TMRL significantly outperforms existing non-temporal Matryoshka approaches and conventional temporal models on both temporal retrieval and Temporal RAG tasks, while also offering efficient inference and deployment capabilities.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Planning, Routing, and Scheduling: Temporal Planning
• Machine Learning: Representation Learning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

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.

Problem

• Temporal Retrieval
• Retrieval-Augmented Generation
• Temporal Information Retrieval
• Time-aware Retrieval
• Temporal Context

Innovation

• Temporal-aware
• Matryoshka embeddings
• Temporal Retrieval
• RAG
• Representation Learning