摘要
The evolution of the internet and its accessibility in the twenty-first century has resulted in a tremendous increase in the use of social media platforms.Some social media sources contribute to the propagation of fake news that has no real validity,but they accumulate over time and begin to appear in the feed of every consumer producing even more ambiguity.To sustain the value of social media,such stories must be distinguished from the true ones.As a result,an automated system is required to save time and money.The classification of fake news and misinformation from social media data corpora is the subject of this research.Several preprocessing and data improvement procedures are used to gather and preprocess two fake news datasets.Deep text features are extracted using word embedding models Word2vec and Global Vectors for Word representation while textual features are extracted using n-gram approaches named Term Frequency-Inverse Document Frequency and Bag of Words from both datasets individually.Bidirectional Encoder Representations from Transformers(BERT)is also employed to derive embedded representations from the input data.Finally,three Machine Learning(ML)and two Deep Learning(DL)algorithms are utilized for fake news classification.BERT also carries out the classification of embedded outcomes generated by it in parallel with the ML and DL models.In terms of overall performance,the DL-based Convolutional Neural Network stands out in the case of the first while BERT performs better in the case of the second dataset.
