Reproduce “A Survey of Generative Information Retrieval” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tzu-Lin Kuo, Tzu-Wei Chiu, Tzung-Sheng Lin, Sheng-Yang Wu, Chao-Wei Huang, Yun-Nung Chen
This paper presents a comprehensive survey of Generative Retrieval (GR), an emerging information retrieval paradigm where a sequence-to-sequence model directly maps user queries to relevant document identifiers. It defines GR's architecture, categorizes indexing and retrieval strategies, evaluates current performance, and outlines future research directions to enhance its scalability and dynamic corpus management.
Key insights
- The design and semantic richness of Document Identifiers (DocIDs) are critical for the generative model's ability to efficiently retrieve relevant documents.
- Generative Retrieval models demonstrate performance degradation as corpus size increases and struggle with dynamic corpora due to high retraining costs and forgetting issues.
- Utilizing synthetic queries for training and employing strategies like Sharpness-Aware Minimization can help mitigate scalability and dynamic corpus management challenges.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.