Sentimental analysis based federated learning privacy detection in fake web recommendations using blockchain model

Recently, a documented increase has been observed in fake news and broadcast of such reports leads to grave danger to individual as well as societal welfare. There's a danger of political collapse and a subsequent devastating loss of public confidence. The overwhelming quantity of news spread online leads towards impractical manual verification and due to the subtle distinctions within language, detecting fake news is an arduous challenge due to the ability to produce coherent and significant. Nowadays advanced neural language models (NLMs) are frequently utilised widespread in sequence generation domains. Additionally, they may be used to create false reviews, which can subsequently be used to target online review platforms and sway consumers' purchasing choices. This research explores the application of blockchain technology and sentiment analysis to create a privacy-focused system for detecting and analyzing fake web recommendations. The input data comprises sentiment-based features extracted from web recommendations. A generative convolutional Bernoulli bayes neural network is employed for the feature extraction and classification. Further, to strengthen network privacy, blockchain technology has been integrated with federated learning. This work offers an experimental analysis of diverse sentiment data-driven fake recommendation datasets, evaluating performance using accuracy, precision, recall, and F-measure metrics. A comprehensive evaluation of effectiveness is performed for each classifier. Results from the classification process indicated that a predictive model could be developed, leveraging tweet data, to distinguish between spam and non-spam content and to determine associated sentiment. The proposed method achieved 99% accuracy, 94% precision, 93% area under the curve, 94% recall, and 96% F-measure.