Multilayer Perceptron Based on Joint Training for Predicting Popularity

摘要

For predictive analysis, Independent features and feature combination are of equal importance, but most models only focus on either independent features or feature combinations. In this paper, we propose a novel deep network model for predictive analysis. It incorporates two components: wide simple feed-forward neural network and MLP (multilayer perceptron) neural network. The wide simple feed-forward neural network is used to generalize to unseen feature combinations, and MLP neural network's aim to select and memorize vital independent features. The Feed-forward & MLP models are jointly trained for the Feed-forward & MLP model, in order to combine the benefits of selection, memorization and generalization. The results from the experiments show the jointly trained Neural Networks model can achieve ideal accuracy.