Leveraging Financial Data Quality Analytics for Strengthening Anti-Money Laundering Compliance and Regulatory Oversight

Abstract

With the rapid development of digital platforms, fake news and misleading information have also been disseminated to a great extent, posing a problem for reliable information verification and automated text classification. In this study, an effective fake news detection and text classification model called Attention-Enhanced Bidirectional Long Short-Term Memory (AE-BiLSTM) network is proposed. The proposed model incorporates an attention mechanism capable of locating and focusing on the most informative words and contextual features in textual data along with its sequential learning ability from BiLSTM. The proposed framework automatically learns complex linguistic patterns and long-range dependencies from the input texts, which is different from traditional machine learning approaches that heavily depend on manual feature extraction. The attention layer boosts the performance of the classification by allocating greater importance to the textual features that are relevant to classification. Experimental tests conducted on the benchmark fake news and text classification data sets show that the proposed AE-BiLSTM model is more capable of improving the accuracy, precision, recall and F1-score compared to conventional deep learning and baseline models. The results show that attention based recurrent structures are an effective solution for automated misinformation detection and strong text classification tasks.