BNnetXtreme: An Enhanced Methodology for Bangla Fake News Detection Online.

In the last couple of years, the government, and the public have shown a lot of interest in fake news on Bangladesh’s fast-growing online news sites, as there have been significant events in various cities due to unjustifiable rumors. But the overall progress in study and innovation in the detection of Bangla fake and misleading news is still not adequate in light of the prospects for policymakers in Bangladesh. In this study, an enhanced methodology named BNnetXtreme is proposed for Bangla fake news detection. Applying both embedding based (i.e. word2vec, Glove, fastText) and transformer-based (i.e. BERT) models, we demonstrate that the proposed BNnetXtreme achieves promising performance in detection of Bangla fake news online. After a further comparative analysis, it is also discovered that BNnetXtreme performed superior to BNnet-one of the state-of-the-art architectures for Bangla fake news detection introduced previously. The BNnetXtreme especially BERT Bangla base model performed with an accuracy score of 91% and an AUC score of 98%. Our proposed BNnetXtreme has been successful in improving the performance by an increase of 1.1% in accuracy score, 5.6% in precision, 1.1% in F1 score, and about 9% in AUC score.