A Novel Tensor Decomposition Networks for Recommender Systems via Adaptive Moment Estimation Method

Abstract The tag, text, and other multiple auxiliary information are acquired in the social networks. The multi-source auxiliary information is heterogeneous information. A large amount of heterogeneous information would result in extremely difficult to analyze. This paper proposes a deep learning network for multi-heterogeneous information processing. The main idea of the our deep learning network is threefold: (1) using tensor decomposition algorithms to process standard encoded data; (2) mining potential factor from heterogeneous data using multi-layer learning networks. Symmetric non-negative potential factor optimization algorithm can effectively predict the missing values of high dimensional sparse data; (3) adaptive moment estimation optimization algorithm is used to replace the traditional first-order optimization algorithm for stochastic gradient descent(SGD) process. The aggregation of heterogeneous data is essentially a minimization problem of multiple parameters. The rate of the model convergence can be improved by adaptive moment estimation in the training $ Fully documented templates are available in the elsarticle package on CTAN. stage. Simultaneously, it can solve the problem of parameter optimization of large-scale data and processing of non-stationary targets. Finally, experimental results on public vaildation datasets are given to verigy the effectiveness of our proposed blend network.