Analyzing digital consumer insights through RoBERTa LLM based sentiment analysis and topic modeling.

Although the classical models like SVM, Naive bayes, random forest, and XGBoost have a non-significantly improved p -value = 0.1 and even more complex models like GRU, Bi-GRU, and BERT have only marginal statistical relevance ( p = 0.08–0.15), the model has highly significant p -value = 4.87 and p = 0.001, which is supported by a Wilcoxon = 3.92 and p = 0.019.

also in 132,142 other papers