Enhancement of Chatbot Services based on Power Supply Terms, Regulations, and Civil Petition History

Overview

The existing chatbot system at Korea Electric Power Corporation (KEPCO) had low utilization due to its overly simplified responses compared to direct human consultations. To overcome this, the project aimed to upgrade the chatbot to improve the automated response rate and provide consistent, standardized response drafts through AI.

Project description

To train the AI model, 1,599 cases of KEPCO's operational FAQ data were extracted. The project followed a structured procedure: 1. Preprocessing training data (604 passages), 2. Generating and verifying queries based on SmartChat, 3. Training and indexing the preprocessed data in SearchGPT/SmartChat, and 4. Developing a demo page for testing.

The main services include an "Intelligent Search Service" for data such as sales operational guidelines and a "Q&A Service" through the AI chatbot to facilitate conversation-based complaint handling.

Key technological features

• SearchGPT: Completed preprocessing of internal data (604 passages), followed by training and indexing to provide a search service integrated with a demo page
• SmartChat: Generated and verified 1,949 Q-A pairs, then applied them to the chatbot solution to provide an automated query generation service via the demo page

Expected outcomes

By integrating the intelligent search module with the existing chatbot, the project saved on the budget required for service quality upgrades. Compared to previous services, the query range has expanded, enabling the system to answer questions that previously had short responses or contained special characters. Furthermore, collecting a wider variety of real user questions and including them in the training dataset is expected to continuously enhance the performance of the chatbot model.

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=29200