To address the issues of data overload and missing multimodal data in information retrieval and recommendation tasks, a visual-modality generation enhanced multimodal knowledge graph attention network (VG-MKGAN) is proposed. Textual, visual and structural triple information are jointly modeled via a multimodal aggregator, and a graph attention mechanism is incorporated to dynamically capture the most relevant neighborhood semantic information between users and items. Furthermore, to tackle the problem that certain entities lack visual modality data, a modality generation strategy based on the visual features of neighboring nodes is designed to effectively complete visual information. Experimental results on the traditional Chinese medicine recommendation dataset (TCM-MMKG), movie dataset (MovieLens) and music dataset (Last-FM) demonstrate that AUC metrics are improved by 2.7% and 3.1% respectively on the TCM-MMKG and Last-FM datasets, and F1 scores are increased by 11.2% and 2.5% correspondingly, compared with those of the best existing baseline model MKGAT.
本文使用1个自建中医药数据集(Traditional Chinese Medicine Multimodal Knowledge Graph, TCM-MMKG)以及两个公开的数据集——电影数据集MovieLens和音乐数据集Last-FM进行实验。其中,MovieLens和Last-FM数据集中的图像数据,使用搜索引擎根据关键词进行爬取。3个数据集的相关统计如表1所示。
HaoX S, ZhuY, AppalarajuS, et al. MixGen: a new multi-modal data augmentation[C]//Proceedings of the 2023 IEEE/CVF winter conference on applications of computer vision workshops. Piscataway, USA: IEEE, 2023:379-389.
[4]
SunR, CaoX Z, ZhaoY, et al. Multi-modal knowledge graphs for recommender systems[C]//Proceedings of the 29th ACM international conference on information & knowledge management. New York, USA: ACM, 2020:1405-1414.
WangH W, ZhangF Z, ZhangM D, et al. Knowledge-aware graph neural networks with label smoothness regularization for recommender systems[C]//Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining. New York, USA: ACM, 2019:968-977.
[8]
KorenY. Factorization meets the neighborhood: a multifaceted collaborative filtering model[C]//Proceedings of the 14th ACM SIGKDD international conference on knowledge discovery and data mining. New York, USA: ACM, 2008:426-434.
[9]
WangH W, ZhangF Z, WangJ L, et al. RippleNet: propagating user preferences on the knowledge graph for recommender systems[C]//Proceedings of the 27th ACM international conference on information and knowledge management. New York, USA: ACM, 2018:417-426.
[10]
ZhangF Z, YuanN J, LianD F, et al. Collaborative knowledge base embedding for recommender systems[C]//Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining. New York, USA: ACM, 2016:353-362.
[11]
ElahiE, AnwarS, Al-kfairyM, et al. Graph attention-based neural collaborative filtering for item-specific recommendation system using knowledge graph[J]. Expert Systems with Applications, 2025,266:126133.
[12]
HuJ, HooiB, HeB S, et al. Modality-independent graph neural networks with global transformers for multimodal recommendation[C]//Proceedings of the thirty-ninth AAAI conference on artificial intelligence. New York, USA: ACM, 2025:11790-11798.
[13]
CuiX H, QuX L, LiD M, et al. MKGCN: multi-modal knowledge graph convolutional network for music recommender systems[J]. Electronics, 2023,12(12):2688.