Design and Implementation of a Platform for Business Intelligence Knowledge Mining and Graph Construction Based on Deep Learning
Zhenghang Li, Xinyu Li, Xinning Lin
Abstract
With the rapid expansion of e-commerce data, traditional business intelligence systems face challenges in handling complex operational scenarios and uncovering deep business insights. This paper aims to construct an intelligent analytics platform integrating deep learning and knowledge graph technologies to enhance the understanding and predictive capabilities of business data. The system adopts a layered architecture design with hardware-software synergy. By leveraging GPU and FPGA accelerators to optimize computational performance, combined with scheduling optimizations at the software layer, it achieves an efficient distributed training framework. The system employs an enhanced Transformer model for multi-granularity temporal feature extraction, integrates business knowledge constraints to improve prediction accuracy, and develops a distributed training framework to optimize computational resource scheduling. Experimental results demonstrate the platform's capability to process 10,000 data records per second, with model inference latency controlled below 20 milliseconds. GPU and FPGA resource utilization rates reach 78% and 75%, respectively. Research indicates that the hardware-software co-optimized integration of deep learning and knowledge graphs effectively enhances the analytical performance and business value of business intelligence systems.