Virality Prediction on Twitter Using Combined CNN and BERT Models

Nowadays, social media serves as a platform for sharing and receiving information. One indicator of the importance of information is its virality. Virality is an interesting topic in the world of Natural Language Processing (NLP), including the prediction of virality and identifying features that influence it. Research in this area has utilized Bidirectional Encoder Representations from Transformers (BERT) models and yielded promising results. Previous studies aimed to predict virality based on specific features. This research aims to predict virality by combining the abilities of Convolutional Neural Network (CNN) and BERT models. The BERT model transforms textual data into numerical and adds new features, while the CNN model excels in numerical classification. Data for this study is sourced from Twitter and in Indonesian language. The proposed model demonstrates improvements compared to the model from previous research. Specifically, it shows a 7.1% increase in precision, a 9.9% increase in recall, a 3.8% increase in accuracy, and an 8.1% increase in F1-Score compared to the previous model. These findings contribute to future research on similar topics, particularly in leveraging multiple models for prediction tasks. Overall, the study provides insights into effectively utilizing the strengths of various models and combining them in prediction tasks.