Optimizing Dense Retrieval Model Training with Hard Negatives

Abstract

Ranking has always been one of the top concerns in information retrieval researches. For decades, the lexical matching signal has dominated the ad-hoc retrieval process, but solely using this signal in retrieval may cause the vocabulary mismatch problem. In recent years, with the development of representation learning techniques, many researchers turn to Dense Retrieval (DR) models for better ranking performance. Although several existing DR models have already obtained promising results, their performance improvement heavily relies on the sampling of training examples. Many effective sampling strategies are not efficient enough for practical usage, and for most of them, there still lacks theoretical analysis in how and why performance improvement happens. To shed light on these research questions, we theoretically investigate different training strategies for DR models and try to explain why hard negative sampling performs better than random sampling. Through the analysis, we also find that there are many potential risks in static hard negative sampling, which is employed by many existing training methods. Therefore, we propose two training strategies named a Stable Training Algorithm for dense Retrieval (STAR) and a query-side training Algorithm for Directly Optimizing Ranking pErformance (ADORE), respectively. STAR improves the stability of DR training process by introducing random negatives. ADORE replaces the widely-adopted static hard negative sampling method with a dynamic one to directly optimize the ranking performance. Experimental results on two publicly available retrieval benchmark datasets show that either strategy gains significant improvements over existing competitive baselines and a combination of them leads to the best performance.

To be published in SIGIR2021

References in corpus (13)

• REALM: Retrieval-Augmented Language Model Pre-Training
• Passage Re-ranking with BERT
• Approximate Nearest Neighbor Negative Contrastive Learning for Dense Text Retrieval
• Document Expansion by Query Prediction
• Large Batch Optimization for Deep Learning: Training BERT in 76 minutes
• Context-Aware Sentence/Passage Term Importance Estimation For First Stage Retrieval
• Sparse, Dense, and Attentional Representations for Text Retrieval
• RocketQA: An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering
• Learning-to-Rank with BERT in TF-Ranking
• Complementing Lexical Retrieval with Semantic Residual Embedding
• RepBERT: Contextualized Text Embeddings for First-Stage Retrieval
• Distilling Dense Representations for Ranking using Tightly-Coupled Teachers
• Overview of the TREC 2021 deep learning track

Cited by in corpus (4)

• Semantic Models for the First-stage Retrieval: A Comprehensive Review
• Lightweight and Direct Document Relevance Optimization for Generative Information Retrieval
• Robust Training of Temporal GNNs using Nearest Neighbours based Hard Negatives
• JurisTCU: A Brazilian Portuguese Information Retrieval Dataset with Query Relevance Judgments