Fig. 5
From: Federated cross-view e-commerce recommendation based on feature rescaling
Schematic of the DSSM for multi-view data modeling. This diagram illustrates the architecture of the DSSM, which integrates multi-view deep neural networks (DNNs) to handle multi-view data in recommendation tasks. The model is structured in three main stages: the input stage, where multiple user and item views are ingested; the representation stage, where deep neural networks process and transform each view into latent semantic vectors; and the matching stage, where the DSSM aligns and matches these vectors based on similarity metrics to generate recommendations. By maximizing the semantic alignment between user views and item views, the model ranks items by similarity, thus enabling more accurate and contextually relevant recommendations. This structure allows the model to capture complex, high-level semantic relationships across different views, enhancing its effectiveness in multi-view and cross-domain recommendation scenarios.
