Enhancing E-commerce Security through Fake News Detection Using Natural Language Processing and Advanced Feature Engineering Technique

E-commerce has simplified customers' lives and offered a range of items, but it has also made them vulnerable to frauds. Fake news on e-commerce platforms threatens trust, brand image, and economic stability. Researchers have shown that contemporary Natural Language Processing (NLP) and machine learning can stop bogus news. However, e-commerce companies still struggle to distinguish phony news from real information. Fast knowledge diffusion can cause financial loss, reputation damage, and customer distrust. Thus, e-commerce false news identification requires robust and trustworthy methods. This investigation will successfully recognize and discriminate fake news. High Feature Extraction uses Word2vec and Term Frequency-Inverse Document Frequency (TF-IDF) to extract features. The optimum feature subset is determined via feature selection utilizing the least absolute shrinkage and selector operator (LASSO). The study involves four phases: Extraction, selection, classification, and data processing are the four steps. To remove raw data, data preparation utilizes stemming, lemmatization, and stop word removal. The suggested method averages model outputs to reduce overfitting and improve prediction stability. DIstilBERT with multi-stacked LSTM is tested on WELFake and ranked by F1 score, sensitivity, accuracy, and specificity. The multi-stacked LSTM distiller has 99.77% accuracy, far greater than the others do. We can use it to detect bogus news. It boosts customer confidence and Internet commerce legitimacy by improving accuracy and consistency.