Adaptive Propagation Mechanism in Deep Graph Neural Networks for Intelligent Retrieval in E-commerce and News Web Platforms

The growth of online platforms has led to the growth of unstructured information, which demands the application of smart systems to identify the contents effectively and efficiently. The absence of the contextuality and flexibility in the conventional search engines stimulates the development of smarter solutions. This study breaks these limitations by proposing the AP-DGNN-HSO model, which should enhance the ability of intelligent search in news and e-commerce websites. It aims at processing complex (textual and behavioral) data using an integrated pipeline. The initial data sources are the RetailRocket (e-commerce) data and the Gossipcop (news) data. They were scaled by Min-Max normalization to make the datasets similar, feature extraction was conducted with the help of the Discrete Orthogonal S-Transform (DOST) in order to grasp significant time and semantic content of the content. These aspects were fed into the Showcase pipeline by the Adaptive Propagation Deep Graph Neural Network (AP-DGNN) that transferred the corresponding data between graphs that are correctly structured. The Hyperbolic Sine Optimizer (HSO) fine-tuned the weights of the model for faster convergence. Results showed good performance able to achieve 99.4% accuracy on RetailRocket, and 99.3% accuracy on Gossipcop, with very low error rates of 0.6% and 0.7% respectively. This systematic procedure represents the mission of demonstrated robust, scalable smart search effectiveness across multiple platforms.