Paper Number

ECIS2025-1685

Paper Type

CRP

Abstract

In recent times, there has been a significant rethinking of enterprise management, which is increasingly based on customer insight. Customer insight management is crucial to facing today’s business challenges. The adoption and application of sentiment analysis (SA) in modern business environments has attracted significant interest in interpreting real-time analyses of customer insights, which offers tremendous opportunities for enhancing decision-making processes. Recent SA methods of pre-trained machine learning and deep learning models cannot perform efficiently for complex linguistic phenomena. Our paper proposed an integrated large language model (LLM) based novel strategy for SA into an enterprise system. Toward this, we used the information systems (IS) computational design science paradigm to develop a two-stage fused-distillation framework of LLM agents from customer insight. We also utilized instruction-following prompts to understand the interpretibility of the fused-distillation framework. Our analysis of three public datasets showed that our framework achieved high precision (acc-95%) in SA and offered the latest LLM-based enterprise-level solutions and methodological advancements.

Author Connect URL

https://authorconnect.aisnet.org/conferences/ECIS2025/papers/ECIS2025-1685

Author Connect Link

Share

COinS
 
Jun 18th, 12:00 AM

LLM-driven Sentiment Analysis for Predicting Customer Insights in Enterprise Systems: a Computational Design Science Approach

In recent times, there has been a significant rethinking of enterprise management, which is increasingly based on customer insight. Customer insight management is crucial to facing today’s business challenges. The adoption and application of sentiment analysis (SA) in modern business environments has attracted significant interest in interpreting real-time analyses of customer insights, which offers tremendous opportunities for enhancing decision-making processes. Recent SA methods of pre-trained machine learning and deep learning models cannot perform efficiently for complex linguistic phenomena. Our paper proposed an integrated large language model (LLM) based novel strategy for SA into an enterprise system. Toward this, we used the information systems (IS) computational design science paradigm to develop a two-stage fused-distillation framework of LLM agents from customer insight. We also utilized instruction-following prompts to understand the interpretibility of the fused-distillation framework. Our analysis of three public datasets showed that our framework achieved high precision (acc-95%) in SA and offered the latest LLM-based enterprise-level solutions and methodological advancements.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.