Generalized Embedding Machines for Recommender Systems

• DOI: 10.1007/s11633-022-1412-6
• Publication Date: 2024-01-12T16:02:34Z
• Authors: Enneng Yang; Xin Xin; Li Shen; Yudong Luo; Guibing Guo
• Keywords: Feature (linguistics); Convolution (computer science)

Factorization machine (FM) is an effective model for feature-based recommendation which utilizes inner product to capture second-order feature interactions. However, one of the major drawbacks FM that it couldn't complex high-order interaction signals. A common solution change function, such as stacking deep neural networks on top FM. In this work, we propose alternative approach signals in embedding level, namely Generalized Embedding Machine (GEM). The used GEM encodes not only information from itself but also other correlated features. Under situation, becomes high-order. Then can incorporate with and even its advanced variants perform More specifically, paper utilize graph convolution (GCN) generate embeddings. We integrate several FM-based models conduct extensive experiments two real-world datasets. results demonstrate significant improvement over corresponding baselines.