The construction industry faces numerous challenges in extracting and interpreting semantic information from CAD floorplans and other construction data. To address this, Graph Neural Networks (GNNs) have emerged as a powerful solution due to their ability to maintain the original structural properties of CAD drawings without rasterization. Identifying structural symbols, such as walls, doors, and windows, is a critical step in generalizing floor plans. This paper investigates GNN methods for improving the classification of multiple structural symbols in CAD floorplans and presents corresponding workflows. We propose an entity-as-node graph representation, study the influence of preprocessing strategies and evaluate different GNN architectures like Graph Attention Network (GAT), GATv2, Generalized Aggregation Networks (GEN), Principal Neighbourhood Aggregation (PNA), and Unified message passing (UniMP) on the CubiCasa5K floorplan dataset. Our results show that the proposed methods outperform state-of-the-art approaches and demonstrate the effectiveness of these methods in CAD floorplan entity classification scenarios.