LangSearch Reranker

Intelligent Ranking for Enhanced Search Results.

LangSearch Reranker is a powerful text-semantic-based ranking model designed to improve the accuracy of search results in search applications and retrieval-augmented generation (RAG) applications. By leveraging deep learning technologies and the Transformer architecture, LangSearch Reranker performs secondary optimization of the initial ranking results, enhancing relevance and overall user experience. Whether you’re using keyword search, vector search, or hybrid search, LangSearch Reranker helps deliver highly relevant results with precise ranking.

Core Features:

Efficient Semantic Ranking LangSearch Reranker utilizes deep semantic matching to perform secondary optimization on initial ranking results (e.g., BM25 or RRF ranking).
- Unlike traditional keyword-based ranking methods, Bocha Semantic Reranker understands the complex semantic relationships between the query and documents, offering results that better align with user intent.
- It refines the ranking by scoring each candidate document more precisely, enhancing the relevance of search results and ensuring users get the most valuable information.
High Performance & Cost Efficiency Based on the Transformer architecture, LangSearch Reranker achieves ranking performance comparable to larger models (such as 280M or 560M parameter models) with only 80M parameters, making it faster, more cost-effective, and more efficient.
- While maintaining high accuracy, it significantly reduces computation and inference costs, making it ideal for large-scale search engines and applications.
- It offers fast response times, ensuring rapid ranking optimization even for large-scale queries.
Versatile for Multiple Search Scenarios Whether you're dealing with keyword search, vector search, or hybrid search, LangSearch Reranker is adaptable and can enhance various search systems.
- In keyword search, LangSearch Reranker improves the accuracy of traditional search methods like BM25.
- In vector search, it refines result relevance by boosting semantic matching with deep learning models, providing better-quality outcomes.
- In hybrid search scenarios, LangSearch Reranker combines the best features of both search methods to provide more accurate, intelligent ranking results.
Intelligent Scoring Mechanism LangSearch Reranker assigns a rerankScore to each candidate document, reflecting the semantic relevance between the query and the document.
- Score Range: The score ranges from 0 to 1, with higher values indicating a stronger semantic match between the document and the query.
- Score Examples:
  - 0.75 ~ 1: The document is highly relevant and answers the question fully, though it might contain additional unrelated text.
  - 0.5 ~ 0.75: The document is relevant but lacks completeness in answering the question.
  - 0.2 ~ 0.5: The document is somewhat relevant, partially answering the query or only addressing certain aspects.
  - 0.1 ~ 0.2: The document is related to the query but only answers a small part of the question.
  - 0 ~ 0.1: The document is not significantly relevant to the query.

Scoring Process:

The scoring process in LangSearch Reranker is based on the semantic match between the query (user's input question) and the content of the document. The model evaluates the semantic relevance of each document and assigns a score based on how well the document answers the query or aligns with the user's intent.

Semantic Relevance Evaluation: LangSearch Reranker analyzes the deep semantic relationship between the query and each document, assessing whether the document effectively answers the user's query or is closely aligned with the query's intent.
Assigning rerankScore: Based on the semantic relevance, LangSearch Reranker assigns a rerankScore to each document. The higher the score, the more relevant the document is to the user's needs.

By improving the initial ranking results with semantic understanding, LangSearch Reranker ensures that the most relevant documents are presented to the user, enhancing the overall search experience.

PreviousLangSearch Database NextWeb Search API

Last updated 14 days ago

Core Features:

Efficient Semantic Ranking LangSearch Reranker utilizes deep semantic matching to perform secondary optimization on initial ranking results (e.g., BM25 or RRF ranking).

Unlike traditional keyword-based ranking methods, Bocha Semantic Reranker understands the complex semantic relationships between the query and documents, offering results that better align with user intent.
It refines the ranking by scoring each candidate document more precisely, enhancing the relevance of search results and ensuring users get the most valuable information.

High Performance & Cost Efficiency Based on the Transformer architecture, LangSearch Reranker achieves ranking performance comparable to larger models (such as 280M or 560M parameter models) with only 80M parameters, making it faster, more cost-effective, and more efficient.

While maintaining high accuracy, it significantly reduces computation and inference costs, making it ideal for large-scale search engines and applications.
It offers fast response times, ensuring rapid ranking optimization even for large-scale queries.

Versatile for Multiple Search Scenarios Whether you're dealing with keyword search, vector search, or hybrid search, LangSearch Reranker is adaptable and can enhance various search systems.

In keyword search, LangSearch Reranker improves the accuracy of traditional search methods like BM25.
In vector search, it refines result relevance by boosting semantic matching with deep learning models, providing better-quality outcomes.
In hybrid search scenarios, LangSearch Reranker combines the best features of both search methods to provide more accurate, intelligent ranking results.

Intelligent Scoring Mechanism LangSearch Reranker assigns a rerankScore to each candidate document, reflecting the semantic relevance between the query and the document.

Score Range: The score ranges from 0 to 1, with higher values indicating a stronger semantic match between the document and the query.
Score Examples:
- 0.75 ~ 1: The document is highly relevant and answers the question fully, though it might contain additional unrelated text.
- 0.5 ~ 0.75: The document is relevant but lacks completeness in answering the question.
- 0.2 ~ 0.5: The document is somewhat relevant, partially answering the query or only addressing certain aspects.
- 0.1 ~ 0.2: The document is related to the query but only answers a small part of the question.
- 0 ~ 0.1: The document is not significantly relevant to the query.

Scoring Process:

Semantic Relevance Evaluation: LangSearch Reranker analyzes the deep semantic relationship between the query and each document, assessing whether the document effectively answers the user's query or is closely aligned with the query's intent.

Assigning rerankScore: Based on the semantic relevance, LangSearch Reranker assigns a rerankScore to each document. The higher the score, the more relevant the document is to the user's needs.

By improving the initial ranking results with semantic understanding, LangSearch Reranker ensures that the most relevant documents are presented to the user, enhancing the overall search experience.