A comparative selection of best activation pair layer in convolution neural network for sentence classification using deep learning model

Abstract

Many natural language processing jobs need sentiment classification of text content. There is an urgent need, particularly with the rise of social media, to extract meaningful information from vast volumes of data upon that Internet utilizing sentiment analysis. We're interested in adopting deep learning models to handle sentiment classification because of the advances that have been made. We present a framework named fastText along with Convolutional Neural Network in this study (CNN). To commence, we use fastText to create words vector representations that will be fed into the CNN. FastText's purpose is to generate a generative model of a word as well as reflect word distance. This one will enable the parameters to also be established at an advantageous CNN point, which will help neural nets to perform much better in this circumstance. Second, we create a CNN architecture that is appropriate for sentiment analysis. We use two sets of convolutional layers along with pooling layers in this design. This is the first time, to our knowledge, that a 9-layer architecture and design model is developed based on fastText as well as CNN was used to assess the sentim ent of phrases. We use the Rectified Linear Unit (ReLU), Normalization, and Dropout techniques to improve the accuracy and generalizability of our model. We put our methodology to the test on a publicly available dataset of movie review extracts with five labels: negative, slightly negative, neutral, moderately positive, and positive. In this dataset, ourReLU pairwise network obtains a test accuracy of 96.4 percent, outperforming existing neural network models.