Online trading increasingly depends on the fusion of heterogeneous data sources to optimize logistics, forecasting, and decision-making. In this study, we developed a multisource sensing and AI-driven framework that integrates physical sensors (radio frequency identification, near field communication, micro-electro mechanical systems, and resistance temperature detectors), environmental sensors (DHT22), and virtual sensors that capture behavioral and sentiment data from social media. A three-layer perception architecture was implemented using a multinode sensor network controlled by Espressif 32 microcontrollers and the Message Queuing Telemetry Transport protocol, achieving low-latency transmission (<100 ms) across a dataset exceeding 10 million records. Advanced AI models, including neural networks and ensemble methods, were applied to fuse IoT, commerce platform, and social media data. The results of this study demonstrated significant improvements in demand forecasting accuracy, fraud detection, and logistics optimization compared with traditional single-source analytics. The optimized routing algorithms increased the on-time shipment rate from 71 to 97%, while the sentiment analysis of more than 30000 monthly mentions provided important market information. The real-time processing capability ensures scalability and resilience, addressing data heterogeneity and computational challenges. This integrated sensing data fusion framework establishes a foundation for more competitive, responsive, and efficient online commerce systems.