An LSTM based Propaganda Detection System for News Articles

Social media has seen a tremendous rise in the past few years. Covering almost all demographics, it has expanded its reach to daily routine and is on the way to becoming an essential aspect of everyday life. Along with social media, the news is also transforming into its digital form. With the internet, it becomes even more readily available, and anyone can create and access it. With this much influence, digital news becomes heavily vulnerable to propaganda. Propaganda is a biased and misleading technique used to promote or publicize some cause, point of view, or perception toward a particular entity. The prime objective of propaganda is to influence public opinion by manipulating information. Digital news has become a popular tool for malicious entities, as it quickly reaches a broad spectrum of people and has prolonged effects. Propaganda in the news can take many forms, such as sensationalized headlines and articles, biased reporting, or false information being presented as fact. It can easily manipulate and mislead people, even if a factual event gets narrated ill-naturedly. Propaganda news is challenging to identify and can have damaging effects. Hence it becomes an important task to identify news that has propaganda. The existing works that deal with propaganda detection lack multi-granularity and suffer from downcast results. An effective technique is proposed in the bi-level propaganda detection system that works on the article and sentence level, integrated together, and is tested experimentally on a standard dataset that achieves maximum accuracy and F1-Score.