Temporal Preference Optimization for Unsupervised Retrieval

2026-06-16

AI Summary

This work addresses the challenge that unsupervised dense retrievers struggle to model temporal relevance, often retrieving semantically related but temporally misaligned documents. To overcome this limitation, the authors propose TPOUR, a novel approach that introduces preference learning into the temporal dimension for the first time. TPOUR leverages Temporal Retrieval Preference Optimization (TRPO) to steer the model toward preferring temporally aligned documents and incorporates interpolatable time embeddings to generalize across unseen time periods. Built upon unsupervised contrastive learning, the method achieves continuous temporal alignment and cross-temporal generalization without requiring labeled data. Experimental results demonstrate that TPOUR significantly outperforms both supervised and unsupervised baselines on time-aware retrieval tasks: it achieves 12.15% and 15.21% relative gains in nDCG@5 on explicit and implicit queries, respectively, while being 72.7× smaller than Qwen-Embedding-8B.

Abstract

Unsupervised dense retrievers offer scalability by learning semantic similarity from unlabeled documents via contrastive learning, but they struggle to capture the temporal relevance, retrieving semantically related but temporally misaligned documents-an important aspect when a document collection spans multiple time periods (e.g., retrieving documents from 2018-2025 for "Who is the president in 2019?" introduces temporal ambiguity). Existing methods rely on supervised training with explicit timestamps, which are not always feasible. We propose TPOUR (Temporal Preference Optimization for Unsupervised Retriever), which uses our novel training method Temporal Retrieval Preference Optimization (TRPO). TRPO reinterprets preference learning in the temporal dimension, guiding the retriever to favor temporally aligned documents. TPOUR further generalizes to unseen time periods via interpolation in a learned time embedding, enabling continuous temporal alignment. Experiments on temporal information retrieval (T-IR), TPOUR outperforms both unsupervised and supervised baselines. Compared to Qwen-Embedding-8B, despite being about 72.7x smaller, TPOUR Contriever improves average nDCG@5 by +4.04 (+12.15%) on explicit and +4.98 (+15.21%) on implicit queries. We provide our code at https://github.com/agwaBom/TPOUR.

Problem

• Temporal Relevance
• Unsupervised Retrieval
• Dense Retrieval
• Time-aware Search
• Temporal Ambiguity

Innovation

• Temporal Preference Optimization
• Unsupervised Retrieval
• Time-aware Embedding
• Contrastive Learning
• Temporal Information Retrieval