El Niño Sentiment Analysis Using Recurrent Neural Network and Convolutional Neural Network Use GloVe

Putrisia, Denada
Lhaksmana, Kemas Muslim

Building of Informatics, Technology and Science
Vol 6 No 1 (2024): June 2024

Sentiment analysis regarding the El Niño climate change is a crucial aspect in understanding public perception and response. This enables deeper identification and understanding of the sentiments evident in online conversations. Sentiment analysis through deep learning approaches using Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) methods is conducted. RNN is a type of artificial neural network designed to process sequential data such as text or time. On the other hand, CNN utilizes convolutional layers to scan text with filters to capture local features like phrases and keywords determining sentiment. Leveraging GloVe representation technique enables the representation of words in numerical vector form capturing semantic relationships among words, facilitating sentiment analysis related to El Niño on social media. The aim of this study is to evaluate the performance of RNN and CNN methods in classifying El Niño-related sentiment with and without GloVe word representation, and to develop a model that can provide accurate and reliable sentiment analysis results. The contribution of this research indicates that the accuracy of sentiment analysis has been improved and can be a significant reference for further research in the field of text analysis and natural language processing (NLP). This study also emphasizes the crucial role of word representation techniques like GloVe in enhancing the performance of deep learning models. The results of the study indicate that the RNN and CNN methods with the utilization of GloVe provide better sentiment classification related to the El Niño issue in social media data, showing that the use of RNN and CNN models with GloVe features perform better compared to not using GloVe features. The use of the RNN algorithm with 80:20 split ratio testing produced an accuracy score of 94.90%, recall of 94.90%, precision of 94.94%, and F1-Score of 94.85%. Meanwhile, the use of the CNN algorithm with 90:10 split ratio testing produced an accuracy score of 94.61%, recall of 93.61%, precision of 94.69%, and F1-Score of 94.58%. This results in the conclusion that sentiment analysis using RNN modeling with GloVe features has better performance than CNN modeling, with an average accuracy rate of 94.90%.