InteractionNN: A Neural Network for Learning Hidden Features in Sparse Prediction

• DOI: 10.24963/ijcai.2019/602
• Publication Date: 2019-07-28T07:46:05Z
• Authors: Xiaowang Zhang; Qiang Gao; Zhiyong Feng
• Keywords: Pooling; Feature (linguistics); Benchmark (surveying)

In this paper, we present a neural network (InteractionNN) for sparse predictive analysis where hidden features of data can be learned by multilevel feature interaction. To characterize interaction features, InteractionNN consists three modules, namely, nonlinear pooling, layer-lossing, and embedding. Nonlinear pooling (NI pooling) is hierarchical structure and, shortcut connection, constructs low-level interactions from basic dense to elementary features. Layer-lossing feed-forward high-level via correlation all layers with target. Moreover, embedding extract which help in reducing our proposed model computational complex. Finally, experiment evaluates on the two benchmark datasets experimental results show that performs better than most state-of-the-art models regression.