# Prompt-RAG: Pioneering Vector Embedding-Free Retrieval-Augmented Generation in Niche Domains, Exemplified by Korean Medicine

**Authors:** Bongsu Kang, Jundong Kim, Tae-Rim Yun, Chang-Eop Kim  
**arXiv:** [2401.11246](https://arxiv.org/abs/2401.11246)  
**Date:** 2024-01-20  
**Summary page:** https://www.summarizepaper.com/en/paper/2401.11246v1/

## Key points

- Prompt-based Retrieval-Augmented Generation (Prompt-RAG) proposed as a novel approach to enhance performance of generative LLMs
- Prompt-RAG operates without the need for embedding vectors
- Comparison of vector embeddings from Korean Medicine (KM) and Conventional Medicine (CM) documents in specialized domains
- KM document embeddings showed stronger correlations with token overlaps and weaker correlations with human-assessed document relatedness compared to CM embeddings
- Evaluation of Prompt-RAG through a Question-Answering (QA) chatbot application
- Results showed that Prompt-RAG outperformed existing models such as ChatGPT and conventional vector embedding-based RAGs in terms of relevance and informativeness
- Challenges include content structuring and response latency
- Advancements in LLMs expected to encourage use of Prompt-RAG in other domains needing RAG methods.

## Summary

<Prompt-RAG, LLMs, niche domains, vector embeddings, Korean Medicine (KM), Conventional Medicine (CM), limitations, generic LLM-based embeddings, specialized domains, token overlaps, human-assessed document relatedness, Question-Answering (QA) chatbot application, relevance, readability, informativeness, ChatGPT, response latency>

The authors propose a novel approach called  (<b>Prompt-based Retrieval-Augmented Generation</b>) to enhance the performance of generative large language models (<b>LLMs</b>) in . Unlike conventional RAG methods that rely on , Prompt-RAG operates without the need for embedding vectors. To demonstrate the limitations of generic LLM-based embeddings in specialized domains, the authors compared vector embeddings from  and  documents. They found that KM document embeddings showed stronger correlations with token overlaps and weaker correlations with human-assessed document relatedness compared to CM embeddings. The performance of Prompt-RAG was evaluated through a <b>Question-Answering (QA) chatbot application</b>, where responses were assessed for relevance, readability and informativeness. The results showed that Prompt-RAG outperformed existing models such as ChatGPT and conventional vector embedding-based RAGs in terms of relevance and informativeness. Despite challenges like content structuring and response latency, the advancements in LLMs are expected to encourage the use of Prompt-RAG as a promising tool for other domains in need of RAG methods. Overall,this study highlights the potential of prompt-based retrieval augmented generation as an effective approach to improve the performance of generative language models in niche domains without relying on embedding vectors.