Retrievers
Cohere Rerank Retriever
The Cohere Rerank Retriever is a specialized retriever that uses Cohere’s reranking capabilities to improve the relevance of retrieved documents. It works by first retrieving documents from a base vector store retriever, then reranking these documents based on their semantic relevance to the query using Cohere’s AI models.
Node Details
- Name: cohereRerankRetriever
- Type: Cohere Rerank Retriever
- Version: 1.0
- Category: Retrievers
Input Parameters
-
Vector Store Retriever (required)
- Type: VectorStoreRetriever
- Description: The base retriever to fetch initial documents from a vector store.
-
Model Name (optional)
- Type: Options
- Default: “rerank-english-v2.0”
-
Options:
- rerank-english-v2.0
- rerank-multilingual-v2.0
- Description: The Cohere model to use for reranking.
-
Query (optional)
- Type: string
- Description: Specific query to retrieve documents. If not provided, the user’s question will be used.
-
Top K (optional)
- Type: number
- Default: Inherits from base retriever, or 4 if not specified
- Description: Number of top results to fetch after reranking.
-
Max Chunks Per Doc (optional)
- Type: number
- Default: 10
- Description: Maximum number of chunks to produce internally from a document.
Outputs
-
Cohere Rerank Retriever
- Type: BaseRetriever
- Description: The configured Cohere Rerank Retriever object.
-
Document
- Type: Document[]
- Description: Array of retrieved and reranked document objects, containing metadata and page content.
-
Text
- Type: string
- Description: Concatenated string of page content from all retrieved and reranked documents.
Credentials
- Credential Name: cohereApi
- Required Parameters: cohereApiKey
How It Works
- The node first initializes a base retriever (usually a vector store retriever).
- It then creates a CohereRerank compressor using the provided API key, model, and parameters.
- A ContextualCompressionRetriever is created, combining the base retriever and the Cohere reranker.
- When queried, it retrieves documents from the base retriever and reranks them using Cohere’s AI.
- The output can be the retriever itself, the reranked documents, or the concatenated text of the documents.
Use Cases
- Improving relevance of document retrieval in question-answering systems.
- Enhancing search results by considering semantic similarity.
- Creating more accurate document summaries by focusing on the most relevant parts.
Notes
- This node requires a Cohere API key to function.
- The effectiveness of the reranking depends on the quality of the initial retrieval and the chosen Cohere model.
- Consider the trade-off between retrieval speed and accuracy when adjusting the Top K and Max Chunks Per Doc parameters.