Hybrid approach of deep feature extraction using BERT– OPCNN & FIAC with customized Bi-LSTM for rumor text classification

Abstract

Rumor data classification in social media seems essential research due to its dependency on digital communications, and this rumor data makes social media unstable. In Natural Language Processing, word embedding, feature extraction, and classification techniques are challenging in rumor detection research. In this research, two phases of feature extraction techniques are involved for performing word embedding and deep feature extraction. In the first phase, Bidirectional Encoder Representations from Transformers with the Optimized Convolutional Neural Network (BERT-OPCNN) techniques are integrated, and in the second phase, the fastText with information gain of an ant colony optimization technique (FIAC) is proposed. Finally, the vectors formed using BERT-OPCNN and the FIAC embedding model are classified using a customized Bi-LSTM. The experiment is computed and compared with existing techniques for balanced and unbalanced datasets. The results evaluation shows that our proposed FIAC embedding with BERT-OPCNN outperforms all other existing techniques using the customized Bi-LSTM classifier.

Keywords

• Rumor detection
• Deep learning
• FastText
• FIAC embedding
• BERT