Retrieval-Augmented Generation Solutions for Typical Application Process Issues

Yudi Zhang

2024

Abstract

Challenges such as the generation of factually incorrect illusions, privacy issues, and outdated information often hinder the practical deployment in the Large Language Model (LLM). Retrieval-Augmented Generation (RAG), which utilizes advanced retrieval technology, is designed to address these issues. RAG will use the embedded vector model to build an external database for the information to be updated and the information of the application field, enhance the user prompt by adding the retrieved relevant data in the context, and retrieve the matched content of the vector library. In this process, how to improve the retrieval efficiency and quality, and how to improve the robustness of the model are the focus of the method discussed in this paper. Gradient Guided Prompt Perturbation (GGPP) uses top k to minimize the distance between the target paragraph embedding vector and query embedding vector while maximizing the distance between original paragraph embedding and query embedding to reduce the influence of perturbation on the model and improve the robustness of the model. Boolean agent RAG setups improve markup efficiency in a language model by incorporating Boolean decision steps where the language model determines whether to query vector databases based on user input. This setting saves a lot of tokens. GenRT is an algorithm that optimizes reordering and truncation strategies to improve efficiency and accuracy in processing long text. Finally, the application of medical question answering system is cited to find the best combination of the retrieval and LLM model in this field.

Paper Citation

in Harvard Style

Zhang Y. (2024). Retrieval-Augmented Generation Solutions for Typical Application Process Issues. In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI; ISBN 978-989-758-713-9, SciTePress, pages 164-167. DOI: 10.5220/0012917400004508

in Bibtex Style

@conference{emiti24,
author={Yudi Zhang},
title={Retrieval-Augmented Generation Solutions for Typical Application Process Issues},
booktitle={Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI},
year={2024},
pages={164-167},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012917400004508},
isbn={978-989-758-713-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI
TI - Retrieval-Augmented Generation Solutions for Typical Application Process Issues
SN - 978-989-758-713-9
AU - Zhang Y.
PY - 2024
SP - 164
EP - 167
DO - 10.5220/0012917400004508
PB - SciTePress