Paper Number
ICIS2025-1601
Paper Type
Short
Abstract
Question-Answering (QA) systems are important in Communities of Practice (e.g., Stack Overflow), by facilitating more effective access to and aggregation of collective knowledge. However, the contemporary QA systems offer standardized answers to similar questions without accounting for users’ diverse knowledge backgrounds, which fail to solve users’ problems. In response, we propose the K-PAG method, which generates personalized answers tailored to the users’ knowledge backgrounds using LLM. Grounded in scaffolding theory and cognitive load theory, the K-PAG identifies the most cognitive-efficient knowledge path that connects users’ existing knowledge concepts to new concepts. The path is then used as a guide to prompt LLMs to generate personalized answers. A cooperative training framework is proposed to reinforce both the path-planning model to generate knowledge paths that better align with the user’s knowledge background and the LLM’s ability to follow knowledge paths during answer generation. Experiment shows K-PAG outperforms benchmarks on generating personalized answers.
Recommended Citation
Jin, Bei; Cheng, Tan; Xu, Yunjie; and Tan, Chee-Wee, "Personalized Answer Generation Based on Individual Knowledge Background Using Large Language Model" (2025). ICIS 2025 Proceedings. 6.
https://aisel.aisnet.org/icis2025/learn_curricula/learn_curricula/6
Personalized Answer Generation Based on Individual Knowledge Background Using Large Language Model
Question-Answering (QA) systems are important in Communities of Practice (e.g., Stack Overflow), by facilitating more effective access to and aggregation of collective knowledge. However, the contemporary QA systems offer standardized answers to similar questions without accounting for users’ diverse knowledge backgrounds, which fail to solve users’ problems. In response, we propose the K-PAG method, which generates personalized answers tailored to the users’ knowledge backgrounds using LLM. Grounded in scaffolding theory and cognitive load theory, the K-PAG identifies the most cognitive-efficient knowledge path that connects users’ existing knowledge concepts to new concepts. The path is then used as a guide to prompt LLMs to generate personalized answers. A cooperative training framework is proposed to reinforce both the path-planning model to generate knowledge paths that better align with the user’s knowledge background and the LLM’s ability to follow knowledge paths during answer generation. Experiment shows K-PAG outperforms benchmarks on generating personalized answers.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
24-Learning