Retrieval Test/Citation
Last updated
Last updated
The Dify Knowledge Base provides a text retrieval testing feature to debug the recall effects under different retrieval methods and parameter configurations. You can enter common user questions in the Source Text input box, click Test, and view the recall results in the Recalled Paragraph section on the right. The Recent Queries section allows you to view the history of query records; if the knowledge base is linked to an application, queries triggered from within the application can also be viewed here.
Clicking the icon in the upper right corner of the source text input box allows you to change the retrieval method and specific parameters of the current knowledge base. Changes will only take effect during the recall testing process. After completing the recall test and confirming changes to the retrieval parameters of the knowledge base, you need to make changes in Knowledge Base Settings > Retrieval Settings.
Suggested Steps for Recall Testing:
Design and organize test cases/test question sets covering common user questions.
Choose an appropriate retrieval strategy: vector search/full-text search/hybrid search. For the pros and cons of different retrieval methods, please refer to the extended reading Retrieval-Augmented Generation (RAG).
Debug the number of recall segments (TopK) and the recall score threshold (Score). Choose appropriate parameter combinations based on the application scenario, including the quality of the documents themselves.
How to Configure TopK Value and Recall Threshold (Score)
TopK represents the maximum number of recall segments when sorted in descending order of similarity scores. A smaller TopK value will recall fewer segments, which may result in incomplete recall of relevant texts; a larger TopK value will recall more segments, which may result in recalling segments with lower semantic relevance, reducing the quality of LLM responses.
The recall threshold (Score) represents the minimum similarity score allowed for recall segments. A smaller recall score will recall more segments, which may result in recalling less relevant segments; a larger recall score threshold will recall fewer segments, and if too large, may result in missing relevant segments.
When testing the knowledge base effect within the application, you can go to Workspace -- Add Function -- Citation Attribution to enable the citation attribution feature.
After enabling the feature, when the large language model responds to a question by citing content from the knowledge base, you can view specific citation paragraph information below the response content, including original segment text, segment number, matching degree, etc. Clicking Link to Knowledge above the cited segment allows quick access to the segment list in the knowledge base, facilitating developers in debugging and editing.
