MobyDeep: A lightweight CNN architecture to configure models for text classification

Nowadays, trends in deep learning for text classification are addressed to create complex models deal with huge datasets. Deeper usually based on cutting edge neural network architectures, achieving good results general but demanding better hardware than shallow ones. In this work, a new Convolutional Neural Network (CNN) architecture (MobyDeep) tasks is proposed. Designed as configurable tool, resultant (MobyNets) able manage big corpora sizes under low computational costs. To achieve those milestones, the was conceived produce lightweight models, having their internal layers proposed convolutional block. That block designed and customized by adapting ideas from image processing, helping squeezing model reduce The also residual network, covering functions extending up 28 layers. Moreover, middle were optimized connections, remove fully connected top resulting Fully CNN. Corpus chosen recent literature, aiming define real scenarios when comparing configured MobyDeep other state-of the-art works. Thus, three 8, 16 respectively, offering competitive accuracy results.