Skip to main content

Retrieval Testing

The Retrieval Testing module in the Knowledge Base is designed to evaluate the quality of the chunks produced based on a given search query. This module allows users to input a search query, set various parameters, and retrieve and rank relevant chunks. The tool provides flexibility in testing and refining the retrieval process to ensure optimal results.

Retrieval Testing Components​

1. Search Query​

The search query is the input text that the user wants to test. This query will be used to search through the Knowledge Base to find relevant chunks of information.

2. Similarity Threshold​

The similarity threshold sets the minimum level of similarity required for a chunk to be considered relevant to the search query. Adjust the similarity threshold to filter out less relevant chunks. A higher value means stricter matching criteria(e.g., 0.0 to 1.0).

3. Top K Results​

This parameter determines the number of top results to be retrieved based on the initial similarity score. Specify the number of top results to retrieve. This helps in limiting the output to the most relevant chunks.

4. Re-rank Model​

The re-rank model is used to further refine the ranking of the retrieved chunks based on more sophisticated algorithms. Choose a re-rank model to reorder the initially retrieved top k results for improved relevance.

5. Re-rank Model Top K Results​

This parameter sets the number of top results to be re-ranked using the selected re-rank model. Specify the number of top results to be re-ranked. This helps in applying the re-rank model to a subset of the initial results for detailed analysis.

6. Get Chunk​

The Get Chunk will initiates the retrieval and ranking process based on the provided parameters. Click to retrieve and display the relevant chunks based on the search query and specified parameters.

Steps to Use Retrieval Testing​

  1. Enter the Search Query: In the search query field, type the text you want to search for in the Knowledge Base.
  2. Set the Similarity Threshold: Adjust the similarity threshold to define the minimum relevance score for chunks to be considered.
  3. Define Top K Results: Specify the number of top results to retrieve based on the initial similarity score.
  4. Select Re-rank Model: Choose a re-rank model from the dropdown menu to further refine the ranking of the retrieved chunks.
  5. Set Re-rank Model Top K Results: Define the number of top results to be re-ranked using the selected model.
  6. Click the Get Chunk Button: Initiate the retrieval process to display the relevant chunks based on the specified parameters.

Example​

  1. Search Query: "AI in healthcare"
  2. Similarity Threshold: 0.8
  3. Top K Results: 10
  4. Re-rank Model: BERT Re-ranker
  5. Re-rank Model Top K Results: 5
  6. Get Chunk Button: Click to retrieve the chunks

This will return the top 10 chunks with a similarity score of at least 0.8 to the search query "AI in healthcare," re-rank the top 5 using the BERT Re-ranker, and display the results.

By using the Retrieval Testing module effectively, users can fine-tune the search and retrieval process, ensuring that the most relevant information is surfaced and ranked appropriately.