Q&A Chatbot for Library Archives and Services

Overview

The diversification of library services and expansion of data have made it difficult for users to obtain desired information immediately, and people have faced limitations in addressing this. Therefore, this project aims to utilize generative AI to learn library data and provides a service that allows users to easily find the information they seek.

Project description

The project utilized a PDF Annotation Tool to extract Q&A sets, followed by data preprocessing and a rigorous review of labeled data for model training. This training incorporated 286 items from the National Library of Korea, including FAQs, user guides, procedure manuals, and promotional materials. To optimize performance, the system trained a retrieval engine on MaumGPT (13B parameters) and implemented two conventional search engines for comparative analysis. Ultimately, the service developed a GPT-based chatbot designed to provide precise answers by leveraging website FAQs and documents from the National Knowledge Resource Sharing Service.

Key technological features

• LLM Training Dataset: Collecting and preprocessing training data by extracting and converting Q&A sets using the PDF Annotation Tool
• Retrieval Engine Implementation: Completed implementation of a search engine using documents from the National Knowledge Resource Sharing Service and frequently asked questions from websites as QA data
• Integration of the Retrieval Engine with the MaumGPT model (13B parameters)

Expected outcomes

The project developed a high-precision GPT model using a curated dataset of 286 Q&A pairs. Comparative analysis results are as follows:

• Retrieval + MaumGPT: 89% accuracy (97/108 queries).
• General MaumGPT: 12% accuracy (6/50 queries).

The system enhances user convenience by effectively translating general inquiries into structured library service information.

Launch year: 2023

Attribution

Case study courtesy of the National Information Society Agency, Republic of Korea, Apolitical's Content Partner. Sourced from Use Cases of Public AI Service, Vol.2, https://eng.nia.or.kr/site/nia_eng/ex/bbs/View.do?cbIdx=31975&bcIdx=29182&parentSeq=29184