Paper Type
Short
Paper Number
PACIS2026-1487
Description
Live stream commerce has emerged as a critical digital channel for customer engagement and online sales. While prior research has examined individual behaviors such as sentiment or purchase intention, limited attention has been paid to the dynamic role structure within live stream ecosystems. Participants continuously shift functional roles during live interactions, creating challenges for real-time analytics and managerial decision-making. This study proposes a KOXs (Key Opinion X) role framework, conceptualizing five core participant roles, and develops a hybrid AI analytics framework for dynamic role identification and behavioral inference. Using multimodal interaction data from Facebook, Instagram, and TikTok—including speech-to-text transcripts, user comments, and temporal signals—this research integrates text embedding techniques, SVM, and LLMs for real-time role classification and evolution tracking. Contributions include a multi-role conceptualization of live stream commerce participation, a hybrid AI-driven analytics framework, and practical implications for real-time marketing decision support.
Recommended Citation
Chou, Tung-Hsiang; Chou, Chao-Chun; Hsieh, Cheng-Yu; Lin, Bei-Ting; and LIAO, TZU-YEN, "Dynamic Role Identification and Behavioral Inference of KOXs in Live Stream Commerce: A Hybrid AI Analytics Framework" (2026). PACIS 2026 Proceedings. 6.
https://aisel.aisnet.org/pacis2026/data_analtyics/data_anltics/6
Dynamic Role Identification and Behavioral Inference of KOXs in Live Stream Commerce: A Hybrid AI Analytics Framework
Live stream commerce has emerged as a critical digital channel for customer engagement and online sales. While prior research has examined individual behaviors such as sentiment or purchase intention, limited attention has been paid to the dynamic role structure within live stream ecosystems. Participants continuously shift functional roles during live interactions, creating challenges for real-time analytics and managerial decision-making. This study proposes a KOXs (Key Opinion X) role framework, conceptualizing five core participant roles, and develops a hybrid AI analytics framework for dynamic role identification and behavioral inference. Using multimodal interaction data from Facebook, Instagram, and TikTok—including speech-to-text transcripts, user comments, and temporal signals—this research integrates text embedding techniques, SVM, and LLMs for real-time role classification and evolution tracking. Contributions include a multi-role conceptualization of live stream commerce participation, a hybrid AI-driven analytics framework, and practical implications for real-time marketing decision support.
Comments
05-DataAnalytics