Developing Sentiment Analysis of Indonesian Social Media Based on Convolutional Neural Network for Smarter Society

Abstract

The need for mining the public's opinion on specific national development issues can be attained by sentiment analysis. There has been much research on sentiment analysis in Indonesian, but there is room for improvement. In this study, we performed experimentation by comparing previously tested convolutional neural network (CNN) architecture with our proposal utilizing a pre-trained word-embedded model: Word2Vec and GloVe that proved to be having better accuracy. The experiment used the SmSA dataset from IndoBERTweet of 11000 sentences. Utilization of the pretrained Word2Vec and GloVe model that was created from 16.2 million BRIN monolingual text corpus could represent a text in a better way. The word-embedding model can capture the semantic and syntactic meaning of a word so that they can conceive knowledge that may be used to raise classification performances. The experimental results show a better F1 score value of 80.06%, which increases by approximately 2.24%.