Agentic data science with the Data Agent Kit
cloud.google.com·8h ago
AI Research
arXiv CS.AI·25d ago·5 min read
Quanta: A Self-Contained Python Library for Hybrid Retrieval over Quantised Embeddings, Lexical Indexes, and Knowledge Graphs
TL;DR
Quanta is an open-source Python library that integrates dense vector search, full-text retrieval, and knowledge graph traversal into a unified retrieval API. It enhances hybrid retrieval systems by using weighted reciprocal rank fusion and treating graph traversal as a candidate expander rather than a relevance scorer.
✦ Why It Matters
Start using Quanta today to simplify your hybrid retrieval implementations and improve search accuracy.
Key Takeaways
- Utilize Quanta to unify multiple retrieval methods in your projects.
- Implement weighted reciprocal rank fusion for better signal integration.
- Leverage graph traversal to expand candidate pools effectively.
- Reduce integration complexity by using a single retrieval API.
How It Works
Quanta integrates three retrieval techniques: it uses 4-bit quantised embeddings for efficient dense vector search, BM25 for robust full-text retrieval, and knowledge graph traversal to expand the candidate pool. The library employs weighted reciprocal rank fusion to combine results, which allows for a more nuanced integration of different retrieval signals without the drawbacks of score normalization.
This design choice ensures that the retrieval process is more adaptable to varying query contexts.