User Sentiment Analysis Towards Online Loan Services Based on Social Media Data X Using the Bert Method

Abstract

The exponential growth of digital media necessitates efficient automated news organization. This research proposes a hybrid classification system integrating K-Means Clustering for automatic data labeling and Long Short-Term Memory (LSTM) for deep learning-based classification. The methodology involves preprocessing unlabeled news datasets and extracting features via TF-IDF. Using K-Means, the data was grouped into six distinct categories: Politics, Economics, Sports, Entertainment, Technology, and Others, validated by Elbow and Silhouette analysis. Subsequently, an LSTM architecture comprising Embedding, LSTM, and Dense layers was trained on this labeled data using a 70:15:15 split. Experimental results demonstrated superior performance, achieving a testing accuracy of 98.19% with high precision, recall, and F1-Scores across all categories. This study concludes that the hybrid K-Means and LSTM approach effectively handles unlabeled datasets, offering a robust solution for automated news content management.