RocketQA: An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering

Abstract

In open-domain question answering, dense passage retrieval has become a new paradigm to retrieve relevant passages for finding answers. Typically, the dual-encoder architecture is adopted to learn dense representations of questions and passages for semantic matching. However, it is difficult to effectively train a dual-encoder due to the challenges including the discrepancy between training and inference, the existence of unlabeled positives and limited training data. To address these challenges, we propose an optimized training approach, called RocketQA, to improving dense passage retrieval. We make three major technical contributions in RocketQA, namely cross-batch negatives, denoised hard negatives and data augmentation. The experiment results show that RocketQA significantly outperforms previous state-of-the-art models on both MSMARCO and Natural Questions. We also conduct extensive experiments to examine the effectiveness of the three strategies in RocketQA. Besides, we demonstrate that the performance of end-to-end QA can be improved based on our RocketQA retriever.

arXiv commentNAACL 2021

References in corpus (11)

• Distilling the Knowledge in a Neural Network
• Training Deep Nets with Sublinear Memory Cost
• Deeper Text Understanding for IR with Contextual Neural Language Modeling
• Passage Re-ranking with BERT
• Efficient Natural Language Response Suggestion for Smart Reply
• Document Expansion by Query Prediction
• Multi-Stage Document Ranking with BERT
• Pre-training Tasks for Embedding-based Large-scale Retrieval
• Knowledge Guided Text Retrieval and Reading for Open Domain Question Answering
• Sparse, Dense, and Attentional Representations for Text Retrieval
• Generation-Augmented Retrieval for Open-domain Question Answering

Cited by in corpus (21)

• Semantic Models for the First-stage Retrieval: A Comprehensive Review
• A Few Brief Notes on DeepImpact, COIL, and a Conceptual Framework for Information Retrieval Techniques
• The Expando-Mono-Duo Design Pattern for Text Ranking with Pretrained Sequence-to-Sequence Models
• BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Models
• mMARCO: A Multilingual Version of the MS MARCO Passage Ranking Dataset
• Optimizing Dense Retrieval Model Training with Hard Negatives
• Condenser: a Pre-training Architecture for Dense Retrieval
• Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup
• UnitedQA: A Hybrid Approach for Open Domain Question Answering
• Pseudo Relevance Feedback with Deep Language Models and Dense Retrievers: Successes and Pitfalls
• SPLADE: Sparse Lexical and Expansion Model for First Stage Ranking
• Matching-oriented Product Quantization For Ad-hoc Retrieval
• Salient Phrase Aware Dense Retrieval: Can a Dense Retriever Imitate a Sparse One?
• Dealing with Typos for BERT-based Passage Retrieval and Ranking
• RocketQAv2: A Joint Training Method for Dense Passage Retrieval and Passage Re-ranking
• Simple Entity-Centric Questions Challenge Dense Retrievers
• Ember: No-Code Context Enrichment via Similarity-Based Keyless Joins
• Unsupervised Document Expansion for Information Retrieval with Stochastic Text Generation
• Jointly Optimizing Query Encoder and Product Quantization to Improve Retrieval Performance
• Ultra-High Dimensional Sparse Representations with Binarization for Efficient Text Retrieval
• Cross-Batch Negative Sampling for Training Two-Tower Recommenders