Title: An Ensembled Latent Factor Model via Differential Evolution and Gradient Descent Optimization

Abstract: In numerous big data contexts, high-dimensional and incomplete (HDI) data are ubiquitous. Latent factor models represent a standard technique for representation learning, effectively extracting meaningful latent factors from such complex datasets. However, traditional latent factor models typically depend exclusively on gradient descent for optimization. This singular reliance can result in representations that are both biased and incomplete, especially when analyzing heterogeneous HDI data. To address these limitations, this research introduces the Ensembled Latent Factor Model via Differential Evolution and Gradient Descent Optimization (ELFM-DEGDO). The framework features a dual-component design: first, it constructs two distinct latent factor models, optimizing one through differential evolution and the other via gradient descent; second, it integrates these two models using a tailored self-adaptive weighting mechanism to synergistically combine their respective strengths. By harnessing the complementary benefits of both optimization strategies, ELFM-DEGDO generates more robust and unbiased representations for HDI data. Experimental evaluations across three HDI datasets demonstrate that ELFM-DEGDO consistently outperforms several existing latent factor models.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC