Implementation of NLP Election News Classification Using LSTM Algorithm

Abstract

This study examined the application of a Long Short-Term Memory (LSTM)-based text classification method to categorize election news according to presidential and vice-presidential candidate entities. The core problem addressed was the lack of an automated classification system capable of identifying political affiliations directly within the vast volume of digital news content. In this research, news data were collected from open-access sources and automatically labeled based on the occurrence of candidate-related keywords. A supervised learning approach was implemented using the LSTM architecture to capture sequential patterns within the news text. The evaluation results demonstrated that the model achieved a validation accuracy of 95.44% and a macro-averaged F1-score of 0.95, indicating strong classification performance across all candidate categories. Furthermore, predictions on test data revealed the model’s consistency and stability in recognizing political entities. This study confirmed the effectiveness of the LSTM-based approach for entity-based election news classification and highlighted its potential for integration into automated media analytics and political discourse monitoring systems.