Text Sentiment Classification Using Knowledge Transfer-Based Matrix Diffractive Riemannian Residual Triangulation Topology Aggregation Model

Text sentiment classification is the process of investigating and categorizing text data based on the sentiment expressed within it, typically into groups such as positive, negative, or neutral. Existing techniques for text sentiment classification often face challenges such as inefficient feature extraction, suboptimal feature selection, overfitting, and high computational complexity. To overcome challenges in text sentiment classification, a comprehensive approach is proposed. The preprocessing phase ensures data integrity and credibility by removing duplicates and unverified reviews, filtering out impure characters, eliminating stop words, and standardizing text through tokenization and punctuation removal. For feature extraction and selection, Term Frequency-Inverse Document Frequency (TF-IDF) coupled with Knowledge Transfer-based Adaptive Differential Evolution (KTADE) optimizes the selection of feature subsets, enhancing the effectiveness of sentiment analysis. Additionally, the Matrix Diffractive Riemannian Residual Triangulation Topology Aggregation (MDRRTTA) neural network model, integrated with a triangulation topology aggregation optimizer, efficiently classifies sentiments, reducing computational complexity and enhancing training speed. This holistic method marks a significant advancement in sentiment analysis methodologies. Proposed model exhibits outstanding performance, achieving remarkable accuracies of 99.56%, 99.35%, and 99.9% for Amazon, Yelp, and Internet Movie Database datasets, respectively.