←back to docs

Rerankers

The reranking algorithm enhances the accuracy of search results from your datasets by applying intelligent secondary sorting. This feature works seamlessly with Retrieval-Augmented Generation (RAG) to ensure the most relevant information appears first in your AI responses.

How Reranking Works

The reranking process adds an intelligent layer between initial retrieval and final results:

Step 1: Initial Retrieval

Your dataset's store performs the initial search and returns a set of potentially relevant records based on your query.

Step 2: Intelligent Reranking

The reranking algorithm analyzes these initial results using advanced relevance scoring, considering factors like:

  • Query-content semantic similarity
  • Context relevance
  • Content quality indicators

Step 3: Optimized Results

The AI receives the reordered results with the most relevant information prioritized, leading to more accurate and helpful responses.

Key Benefits

  • Enhanced Accuracy - Secondary sorting significantly improves the relevance of returned results, reducing irrelevant or low-quality responses.
  • Maintained Performance - The algorithm processes the same limited set of records, so response times remain fast while quality improves.
  • Cost-Effective - This advanced feature is included at no additional cost with your ChatBotKit subscription.

Setup Instructions

Enabling Reranking

  1. Navigate to your Dataset Settings in the ChatBotKit dashboard
  2. Find the "Reranking Algorithm" section
  3. Toggle the feature ON
  4. Configure any available settings based on your use case
  5. Save your changes

Configuration Options

The specific configuration options available depend on your dataset size and store type. Common settings include:

  • Relevance threshold adjustment
  • Result diversity preferences
  • Domain-specific optimization

Best Practices

  • Test with representative queries to evaluate improvement in result quality
  • Monitor performance after enabling to ensure it meets your expectations
  • Provide feedback on results quality to help improve the algorithm

Troubleshooting

Not seeing improved results?

  • Ensure the feature is enabled in your dataset settings
  • Try with different query types to identify optimal use cases
  • Check that your dataset has sufficient content for meaningful reranking

Performance concerns?

  • Monitor response times before and after enabling
  • Contact support if you notice significant slowdowns
storage