Paper Number
ICIS2026-2661
Paper Type
Completed
Abstract
As user growth on creator platforms slows, creators face higher costs to attract audience attention, making content optimization increasingly important. A key challenge is that observed engagement reflects both intrinsic content quality and the platform’s shifting exposure-allocation logic, which changes endogenously with recommendation strategies. We conceptualize this distortion as an information systems design problem and develop an adaptive decision-support artifact for video attractiveness prediction and optimization in dynamic platform environments. The artifact integrates robustness-oriented offline learning with online reinforcement learning. The offline module uses farthest-domain centroid cross-domain matching and a domain-adversarial neural network to distinguish stable content signals from changing recommendation conditions. The online module employs a Base-Model-as-Critic dual Actor–Critic algorithm to recalibrate environmental interpretation and refine multimodal video features. Experiments on real-world platform data show that the artifact outperforms baseline methods in prediction accuracy and optimization stability, while simulations generate actionable content optimization strategies.
Recommended Citation
Shi, Xiaowei; Wang, Cong; and Wu, Peng, "From Prediction to Prescription: Adaptive Video Content Optimization under Evolving Recommendation Regimes" (2026). ICIS 2026 Proceedings. 11.
https://aisel.aisnet.org/icis2026/datamgmt/datamgmt/11
From Prediction to Prescription: Adaptive Video Content Optimization under Evolving Recommendation Regimes
As user growth on creator platforms slows, creators face higher costs to attract audience attention, making content optimization increasingly important. A key challenge is that observed engagement reflects both intrinsic content quality and the platform’s shifting exposure-allocation logic, which changes endogenously with recommendation strategies. We conceptualize this distortion as an information systems design problem and develop an adaptive decision-support artifact for video attractiveness prediction and optimization in dynamic platform environments. The artifact integrates robustness-oriented offline learning with online reinforcement learning. The offline module uses farthest-domain centroid cross-domain matching and a domain-adversarial neural network to distinguish stable content signals from changing recommendation conditions. The online module employs a Base-Model-as-Critic dual Actor–Critic algorithm to recalibrate environmental interpretation and refine multimodal video features. Experiments on real-world platform data show that the artifact outperforms baseline methods in prediction accuracy and optimization stability, while simulations generate actionable content optimization strategies.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
07_Data_management