Most traditional goods recommendation algorithm have difficulty in capturing higher-order information about interaction, while heterogeneous graph neural networks can capture complex topological information and preserve the heterogeneous information of attributes in heterogeneous network, thus extracting richer structural and semantic representation information and improving goods recommendation. Therefore, we proposed a multi-modal personalized goods recommendation based on Graph Attention-Enhanced Graph Neural Network (GAGN) by combining graph neural network related method and using the interaction behavioral data of user and goods. GAGN mines user and goods node representation on user-goods interaction graph through long-term user interest model and goods model; Text information extractor to learn user and goods reviews, goods descriptions and information representations; Extracting users’ short-term preference representations by model their short-term interests and interlinking multi-modal information such as node representation, review representation, descriptions of goods, and users’ short-term preferences through a graph attention-enhanced aggregator to achieve better information fusion. Finally, the proposed algorithm is fully experimented on three real data sets, thus demonstrating the superiority of the proposed algorithm over others benchmark model.