Han Xiaofei, Dou Xin
Business College, California State University, Long Beach, CA, United States.
School of Business and Management, Shanghai International Studies University, Shanghai, China.
Front Neurorobot. 2025 Jun 18;19:1587973. doi: 10.3389/fnbot.2025.1587973. eCollection 2025.
In common graph neural network (GNN), although incorporating social network information effectively utilizes interactions between users, it often overlooks the deeper semantic relationships between items and fails to integrate visual and textual feature information. This limitation can restrict the diversity and accuracy of recommendation results. To address this, the present study combines knowledge graph, GNN, and multimodal information to enhance feature representations of both users and items. The inclusion of knowledge graph not only provides a better understanding of the underlying logic behind user interests and preferences but also aids in addressing the cold-start problem for new users and items. Moreover, in improving recommendation accuracy, visual and textual features of items are incorporated as supplementary information. Therefore, a user recommendation model is proposed that integrates hierarchical graph attention network with multimodal knowledge graph. The model consists of four key components: a collaborative knowledge graph neural layer, an image feature extraction layer, a text feature extraction layer, and a prediction layer. The first three layers extract user and item features, and the recommendation is completed in the prediction layer. Experimental results based on two public datasets demonstrate that the proposed model significantly outperforms existing recommendation methods in terms of recommendation performance.
在常见的图神经网络（GNN）中，尽管整合社交网络信息能有效利用用户之间的交互，但它常常忽略物品之间更深层次的语义关系，并且无法整合视觉和文本特征信息。这种局限性可能会限制推荐结果的多样性和准确性。为了解决这个问题，本研究将知识图谱、GNN和多模态信息相结合，以增强用户和物品的特征表示。知识图谱的纳入不仅有助于更好地理解用户兴趣和偏好背后的潜在逻辑，还有助于解决新用户和新物品的冷启动问题。此外，在提高推荐准确性方面，物品的视觉和文本特征被作为补充信息纳入。因此，提出了一种将分层图注意力网络与多模态知识图谱相结合的用户推荐模型。该模型由四个关键组件组成：协作知识图谱神经层、图像特征提取层、文本特征提取层和预测层。前三层提取用户和物品特征，推荐在预测层完成。基于两个公共数据集的实验结果表明，所提出的模型在推荐性能方面显著优于现有推荐方法。
