Semantic-based padding in convolutional neural networks for improving the performance in natural language processing. A case of study in sentiment analysis

In this work, a methodology for applying semantic-based padding in Convolutional Neural Networks for Natural Language Processing tasks is proposed. Semantic-based padding takes advantage of the unused space required for having a fixed-size input matrix in a Convolutional Network effectively, using words present in the sentence. The methodology proposed has been evaluated intensively in Sentiment Analysis tasks using a variety of word embeddings. In all the experimentation carried out the proposed semantic-based padding improved the results achieved when no padding strategy is applied. Moreover, when the model used a pre-trained word embeddings, the performance of the state of the art has been surpassed. (C) 2019 Published by Elsevier B.V.