A Novel Text Separable Attention-Based Sentiment Classification and Rating Network with Hybrid Feature Extraction

Sentiment classification involves analyzing text to determine emotional tone — positive, negative, or neutral. Challenges include interpreting context, managing diverse expressions, and handling language complexities, with existing models also struggling to scale to large text volumes. To address these issues, this research introduces the novel Text Separable attention selective subject Encoder Convolution network with Genghis Khan Shark Optimization (TSEC-GKSO) algorithm, which classifies features into sentiment categories and predicts ratings from one to five stars. The Adaptive Union Method (AUM) selects features by combining top-ranked features through union and applying intersection on remaining subsets, aiming to minimize feature space dimensions while retaining both top-ranked and common features. The TSEC-GKSO method achieved accuracies of 99.80% on RES14, 99.41% on RES15, and 99.54% on both Lap14 and IMDB datasets. It effectively addresses sentiment analysis challenges with accurate classification and efficient large dataset handling.