Rerank Based on Summary

[medcl]

Description:

In the current search process, after receiving the search results, we aim to enhance relevance by reranking them based on a high-level summary, such as the title or meta-description. Here's the proposed approach:

1. Initial Judgement with LLM:

   • Upon obtaining the search results, utilize a Large Language Model (LLM) to assess the relevance of each result based on its summary (e.g., title, meta description).
   • The LLM will exclude results that are obviously irrelevant or incorrect, improving the accuracy of the initial ranking without fetching the full document.

2. Fetch Document Content for Summarization:

   • For the top-ranked results that pass the LLM's filtering, fetch the full document content.
   • Apply further summarization to provide detailed insights for deeper analysis or final decision-making.