A sentiment analysis study for twitter using the various model of convolutional neural network

Abstract

Sentiment analysis about natural language processing had been developed by various methods such as machine learning. In Indonesia, fewer studies on sentiment analysis have been studied using machine learning with specific datasets. The use of machine learning results in low accuracy that is applied in a wider scope. Therefore, the deep learning method developing to improve the accuracy of sentiment analysis. This paper presents different configuration parameters based on deep learning using the Convolutional Neural Network (CNN) algorithm to improve accuracy performance. The CNN models are presented by various parameters such as the number of convolutional layers, number of filters, and filter size to analyze the model’s performance. Indonesian-Sentiment-Analysis-Dataset which consists of 10.806 tweets has been used with the Word2Vec model for Indonesian as a word vector representation. The CNN models are trained on 80% of the dataset and tested on the remaining 20% of the dataset. The proposed CNN models’ results are compared with machine learning algorithms such as SVM, KNN, and SGD. The CNN models performed better than machine learning and got the best accuracy of 81.4% for general sentiment analysis in Indonesian.