A Multi-Task Learning Approach for News Classification and Number of Reader Prediction

The large volume of online news content presents a challenge in effectively managing and organizing information, especially regarding enhancing literacy rates in Indonesia. As the amount of news articles continues to grow, there is a need for a robust system that can categorize news and predict the number of readers to assess its impact on literacy. This study introduces a Multi-Task Learning (MTL) approach, utilizing data from online websites to simultaneously address news classification and reader prediction tasks. Cross-entropy loss is applied in the model to handle the class imbalance issue. The research compares two performance MTL architectures, the Dense architecture and the CNN architecture. The experiments assess the models' abilities to classify news and predict reader numbers. The results show that the Dense architecture outperforms the CNN architecture, achieving 94% accuracy and a 99% AUC-ROC score, whereas the CNN model achieved 91% accuracy and a 98% AUC-ROC score. This study highlights the effectiveness of the Dense architecture in classifying online news and predicting reader engagement. The findings provide valuable insights for enhancing news sorting systems and could contribute to improving literacy initiatives in Indonesia by offering more accurate predictive models for online news consumption. The results indicate that integrating Multi-Task Learning into news classification systems can enhance content management and offer a deeper understanding of public interaction with news.