DETECT &COMBAT FAKE NEWS &MISINFORMATION ON SOCIAL MEDIA USING MACHINE LEARNING

Abstract

There are benefits and drawbacks to relying only on social media for news updates. It is true that social media platforms facilitate the rapid dissemination of knowledge among individuals. Still, these kinds of websites might be used to spread misinformation-filled, low-quality content—what is known as "fake news." Because malevolent actors may sway readers with misinformation and false news, social media has grown into an integral part of state security, generating needless discussion on matters that are inherently meaningless to society. This leads to a domino effect, public anxiety, and ultimately risks to the security of the state. Early fake news detection is the most important step in saving people's lives from the spread of false information. People unintentionally contribute to the spread of false information by spreading it. Thus, identifying fake news that has been posted on different social media platforms has recently gained a lot of attention as a developing field of study. The detection of false news on the many social media platforms presents new difficulties that render the algorithms now in use obsolete or inefficient. On the other hand, the original propagators of false news seek to disseminate the misinformation by focusing on innocent individuals. Cutting-edge data sensors and deep learning methods hold considerable promise for facilitating the development of useful solutions to address the issue of false news. However, because of data shortages, such remedies often need improved model generalisation in the actual world. In this research, we address the problem of false news identification by introducing a novel approach that uses a committee of classifiers. Neuronal networks have shown to be an excellent tool among others for identifying bogus news on social media. In this study, a deep learning-based methodology has been used to distinguish bogus news from authentic sources. The suggested model has been constructed using an LSTM neural network. To that purpose, we provide a variety of basic models, each of which has been individually trained on sub-corpora with distinct qualities. Specifically, we use multi-label text-type categorization to aid in the formation of an ensemble. The study was carried out using six distinct benchmark datasets. The findings are encouraging and pave the way for more study.