texttron/hyde
HyDE: Precise Zero-Shot Dense Retrieval without Relevance Labels
This project helps information retrieval specialists and researchers improve search accuracy for complex queries without needing to manually label documents for relevance. It takes a natural language query and generates a hypothetical document to help a search engine find more precise results, even in situations where no prior examples of relevant documents exist. The output is a significantly better set of search results for a given query.
573 stars. No commits in the last 6 months.
Use this if you need to perform highly accurate searches on large document collections, especially when dealing with new or specialized topics where human-labeled relevance judgments are scarce or impossible to obtain.
Not ideal if your search needs are simple and can be met with traditional keyword-based search or if you have ample labeled data for training a supervised retrieval model.
Stars
573
Forks
41
Language
Jupyter Notebook
License
—
Category
Last pushed
Dec 06, 2024
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/embeddings/texttron/hyde"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
ewok-core/ewok-paper
Elements of World Knowledge! This repository houses data and code needed to replicate our first...
itrummer/thalamusdb
ThalamusDB: semantic query processing on multimodal data
ArslanKAS/Large-Language-Models-with-Semantic-Search
Explore from keyword search to dense retrieval and reranking, which injects the intelligence of...
Ahren09/SciEvo
A longitudinal dataset for academic literature, including papers, metadata, and citation graphs,...
jzhoubu/vsearch
An Extensible Framework for Retrieval-Augmented LLM Applications: Learning Relevance Beyond...