FANCFIS: ensemble deep learning based features learning with a novel fuzzy approach for sentiment analysis

In the past decade, sentiment analysis has been a popular study area in data mining and natural language processing. In sentiment analysis, deep neural network (DNN) approaches have recently been used in tasks with encouraging results. The usage of deep neural networks for this purpose has grown significantly over the past few years, and the results have been considerable. DNN is showing promise as an approach. Because DNN can independently extract features from data, intermediate representations produced by these networks can also be employed as appropriate features. Therefore, we suggested an ensemble of ConvNeXt and PCFAN in this research to extract both low-level and high-level features. While many deep neural networks can extract different kinds of information because of their unique topologies, we merge features obtained from hybrid neural networks. To improve the entire effectiveness of sentiment analysis by considering their association. A multi-scale classifier called the Hierarchical Multi-scale LSTM (HMLSTM) receives intermediate representations from both introduced deep neural networks. Using the Sentiment140 and SST2 datasets, we simulate and assess the suggested approach. Our suggested model can outperform state-of-the-art approaches, as demonstrated by experimental findings on numerous public benchmark datasets.