A Multi-Modal Big Data Retrieval Method Based on Tensor Decomposition
Multi-modal retrieval is currently a significant research direction in social media retrieval.Large-scale,high-dimensional multi-modal data is characterized by its vast volume and heterogeneous structure.A pressing challenge is how to uncover hidden features while preserving the integrity of multi-modal data to facilitate similarity calculations across different modalities.Traditional retrieval methods,such as matrix decomposition,sub-space methods,and deep learning-based approaches,often employ a two-stage correlation learning process that overlooks high-level semantic information and fails to consider the intrinsic structure of the data,such as sparsity and low-rank properties.The article combines the nonlinear manifold structure of multi-modal data to propose a multi-modal retrieval method that integrates manifold learning.Building upon tensor decomposition,this method processes and analyzes high-dimensional multi-modal data using manifold metrics and distances.The approach is validated through experiments on real datasets,demonstrating its ability to enhance retrieval performance.
NETL
NSTL
万方数据
