Personalized Dialogue Response Generation, which aims to emulate human-like and customized responses, has attracted considerable research interest. Recent advances in this field have integrated external knowledge to enhance models’ language comprehension. However, the generic nature of this external knowledge is inadequate for the generation of personalized responses for a diverse range of users. Moreover, the redundancy of external knowledge diverts models’ attention to other topics. To address these challenges, we propose a novel GNN-based approach, Hierarchical Knowledge Aggregation (HKA), which hierarchically aggregates both external knowledge and knowledge from the dialogue context. Additionally, we introduce a dialogue relation recognition task to enhance the accurate modeling of structural information. The results of our experiments, conducted on the ConvAI2 and PersonalDialog datasets, demonstrate that HKA outperforms existing baselines in generating personalized responses.