Classification of User Expressions on Social Media Using LSTM and GRU Models

Abstract

Social media serves as a platform for sharing information. Through social media, users can interact with others and express their feelings and emotions. Therefore, emotion analysis plays a crucial role in understanding users' conditions regarding various issues and social events. This study aims to compare the performance of emotion classification models in analyzing and identifying users' emotions on social media. The research process includes data preprocessing, training, and model performance evaluation. The dataset used is derived from Twitter social media and is available on Kaggle. It consists of two main columns: text and label, with the latter categorized into six groups. The dataset undergoes several preprocessing techniques to ensure it is ready for model training. The model training process implements the architectures of LSTM and GRU to analyze the emotions contained within the text. The evaluation results show that the model achieves an accuracy of 93% for LSTM and 94% for GRU, indicating that the GRU model slightly outperforms the LSTM in classifying emotions in textual data. This research is expected to contribute to emotion analysis systems based on deep learning.