Sequential recommendation systems are created to predict the next item a user will likely interact with using temporal patterns and contextual information learned from historical user-item interactions. This study evaluates a Dynamic Graph Neural Network for Sequential Recommendation (DGSR) framework that attempts to effectively model user preferences to predict the next BIM command. The proposed method first trains a DGSR model on existing user-command interactions to generate embeddings that capture sequential and structural information. For new, unseen users, a similarity-based weighted average aggregation is introduced to transfer embeddings from the pretrained user nodes to the new nodes based on their initial interactions. These aggregated embeddings are used to infer new user predictions. The model is then partially retrained to boost the new users' representations. This hybrid approach combines the strengths of pretrained embeddings with adaptive retraining, effectively bridging the gap between transductive methods and production environments. Evaluations on a BIM log file dataset show the model's ability to predict the next user commands. The findings have shown the potential of dynamic graph-based recommendation systems in niche domains, providing a robust solution for sequential prediction tasks.     «

Sequential recommendation systems are created to predict the next item a user will likely interact with using temporal patterns and contextual information learned from historical user-item interactions. This study evaluates a Dynamic Graph Neural Network for Sequential Recommendation (DGSR) framework that attempts to effectively model user preferences to predict the next BIM command. The proposed method first trains a DGSR model on existing user-command interactions to generate embeddings that...     »