ABSTRACT
Ongoing research emphasizes the importance of bidirectional training in detecting fake news. Sequential neural networks encode content and social context data, considering unidirectional text sequences. Bidirectional training captures crucial information, improves classification performance, and detects semantic and contextual dependencies in sentences. Three neural network hybrid algorithms – convolutional neural network + recurrent neural networks (CNN+RNN), convolutional neural network + long short-term memory (CNN+LSTM), and convolutional neural network + gated recurrent unit (CNN+GRU) – are examined in this chapter and contrasted with the current algorithms like CNN, RNN, LSTM, and GRU. Each model will be paired with word to vector (Word2Vec) for word embedding and then will have an initial CNN layer. In the suggested work, the effectiveness of each algorithm under the influence of various activation functions and optimizers will be compared. The best performing permutation for fake news detection will then be found by comparing all the results. According to the study’s findings, it is necessary to identify fake news. The experiment results show that for the purpose of ‘fake news prediction’, out of the ensemble models being tested, the ensemble model of CNN + GRU when in conjunction with ‘SGD’ optimizer and ‘SIGMOID’ activation function works best with an accuracy of around 97.8%.
