Debunking multi-lingual social media posts using deep learning

Abstract

Fake news on social media has become a growing concern due to its potential impact on shaping public opinion. The proposed Debunking Multi-Lingual Social Media Posts using Deep Learning (DSMPD) approach offers a promising solution to detect fake news. The DSMPD approach involves creating a dataset of English and Hindi social media posts using web scraping and Natural Language Processing (NLP) techniques. This dataset is then used to train, test, and validate a deep learning-based model that extracts various features, including Embedding from Language Models (ELMo), word and n-gram counts, Term Frequency-Inverse Document Frequency (TF-IDF), sentiments, polarity, and Named Entity Recognition (NER). Based on these features, the model classifies news items into five categories: real, could be real, could be fabricated, fabricated, or dangerously fabricated. To evaluate the performance of the classifiers, the researchers used two datasets comprising over 45,000 articles. Machine learning (ML) algorithms and Deep learning (DL) model are compared to choose the best option for classification and prediction.