Retrieval Modes
When users build AI applications with multiple knowledge bases, Dify’s retrieval strategy will determine which content will be retrieved.
Retrieval Setting
Matches all datasets based on user intent, querying related text fragments from multiple datasets simultaneously. After a re-ranking step, the best results matching the user query are selected from the multi-path query results, requiring a configured Rerank model API. In multi-path retrieval mode, the retriever searches for text content related to the user query across all datasets associated with the application merge the relevant document results from multi-path retrieval and re-ranks the retrieved documents semantically using the Rerank model.
In multi-path retrieval mode, it’s recommended that the Rerank model be configured.
Below is the technical flowchart for the multi-path retrieval mode:
Since multi-path retrieval mode does not rely on the model’s inference capability or dataset descriptions, it can achieve higher-quality retrieval results when retrieving across multiple datasets. Additionally, incorporating a re-ranking step can effectively improve document retrieval effectiveness. Therefore, when creating knowledge base Q&A applications associated with multiple datasets, we recommend configuring the retrieval mode as multi-path retrieval.