An accurate fake news detection approach based on a Levy flight honey badger optimized convolutional neural network model
Dheeraj Kumar Dixit,Amit Bhagat,Dharmendra Dangi
DOI: https://doi.org/10.1002/cpe.7382
2022-11-25
Concurrency and Computation: Practice and Experience
Abstract:Summary People can quickly acquire the news through a variety of sources, including websites, blogs, and social media, among others. The spread of fake news has become easier as a result of the availability of these platforms. Anybody with access to these networks generates and distributes fake news for professional or personal gain. Numerous studies relying on supervised and unsupervised learning techniques are available to address the issue of recognizing fake news. All of those studies, though, have one flaw: they all deliver mostly inaccurate or unmatched results. Poor accuracy is attributed to a variety of factors, including imbalanced datasets, inefficient parameter tuning, poor feature selection, and so on. To tackle these issues, we proposed a novel approach for fake news detection. Initially, the data were obtained from the ISOT dataset and data cleaning is performed. After that, preprocessing is done which includes three major steps such as stemming, stop word removal, and tokenization are carried out. Next to preprocessing, various features that involve name entity recognition‐based features are selected during feature extraction. From this, the short dimensionality features are selected with the help of the ensemble modified independent component analysis model. Finally, the hybrid convolutional neural network‐based Levy flight‐based honey badger algorithm detects fake news. The experiments are simulated using python software with various performance metrics such as accuracy, specificity, sensitivity, precision, and F‐scores to validate the performance of the proposed method. The proposed model offers a precision, recall, and accuracy value of 95%, 97%, and 98% when evaluated with the ISOT dataset. When compared to the existing state‐of‐art methods, the proposed method yielded superior detection results and higher accuracy rates.