Literature Survey: Retrieval Methods

This section contains detailed surveys and analyses of individual papers related to Stage 1 retrieval methods, including dense retrieval, sparse retrieval, hybrid approaches, and hard negative mining strategies.

Papers:

• Hard Negative Mining: A Deep Dive for Building a Unified Mining Library
  • Motivation: The Gap in the Ecosystem
  • How ColBERTv2 Uses Hard Negative Mining
  • Existing Implementations Analyzed
  • Theoretically-Grounded Mining Methods
  • Principled Sampling Methods
  • Synthetic Generation Methods
  • Quantitative Impact of Mining Strategies
  • MUVERA: Multi-Vector Efficiency (Not Mining)
  • Proposed Library Design
  • Detailed Library Implementation Analysis
  • Summary
  • References

Overview

The papers in this section provide in-depth technical analysis of key contributions to the retrieval literature. Each survey includes:

• Problem formulation and mathematical foundations

• Algorithmic innovations with theoretical guarantees

• Empirical results on standard benchmarks

• Practical considerations for deployment

• Connections to other methods in the retrieval-reranking pipeline

Featured Papers

Hard Negative Mining: A Deep Dive for Building a Unified Mining Library

Comprehensive analysis of hard negative mining strategies including ColBERTv2’s training pipeline, theoretically-grounded methods (ANCE, ADORE, RocketQA), and existing implementations (Contrastors, PyLate). Includes proposed API design for a unified mining library analogous to the rerankers library.

Topics Covered

• Dense Retrieval: DPR, ANCE, Contriever, and embedding-based methods

• Sparse Retrieval: BM25, SPLADE, and learned sparse representations

• Hard Negative Mining: Dynamic mining, curriculum learning, false negative handling

• Hybrid Methods: Combining dense and sparse retrieval for robustness

• Pre-training: ICT, contrastive pre-training, and domain adaptation

Contributing

To add a new paper survey to this section:

1. Create a new .rst file following the structure of existing surveys

2. Include: problem statement, core innovation, theoretical analysis, empirical results

3. Add the file to the toctree above

4. Ensure proper citations and links to related papers