Hybrid search is an innovative technique that blends the precision of full-text search with the nuanced understanding of vector search to provide well-rounded search results.
One of the simplest libraries can be found here: https://github.com/Raudaschl/rag-fusion
By executing a single query that incorporates both search and vector query parameters in parallel, and scoring the combined results using the Reciprocal Rank Fusion (RRF) algorithm, hybrid search offers a robust solution for retrieving information.
This approach is particularly advantageous for chat-style and copilot applications, where both the relevance of keyword matches and the conceptual similarity captured by vector embeddings are crucial for high-quality responses.
In practice, a hybrid query involves a search index that houses various data types, enabling a comprehensive search experience. For instance, a user might input a full-text search query to find “historic hotels within walking distance to restaurants and shopping,” and the hybrid system will also consider vector embeddings related to the query to surface the most relevant results.
This could include understanding the context behind phrases without relying on exact keyword matches, thanks to the multi-lingual capabilities of vector search.
The value of hybrid search is underscored by its ability to fine-tune results using existing functionality such as filtering, faceting, semantic ranking, and without the need for explicit sorting, which could otherwise disrupt the relevance-ranked output.
It’s a technique that harnesses the best of both worlds, ensuring users receive information that is both accurate to their specific query and broadly relevant to the topic at hand.
Such capability aligns with the goals of Veso AI, as it aids in enhancing the efficiency and efficacy of search functions within organizational tools like Veso Nexus, ensuring that team members can access the most pertinent information quickly and seamlessly.