E-commerce platforms have complex and dynamically changing webpage structures. Traditional web crawlers and rule-based information extraction methods struggle to adapt to these challenges, resulting in high maintenance costs and low extraction efficiency. To address this issue, this paper proposes a Graph Neural Network (GNN)-based web information extraction algorithm, WebGCN, which effectively leverages the HTML structure by integrating it into the web document representation and incorporates graph attention, sparse attention, and local attention mechanisms to reduce global computational complexity while enhancing extraction accuracy. Experiments on multiple real-world e-commerce datasets demonstrate that WebGCN achieves state-of-the-art performance.