Comprehensive Analysis on Depression using social media: A Bioinspired Algorithm Analysis

Abstract

This article compares different sentiment analysis models, using bio-inspired optimization and fuzzy neurocomputing as a comparison tool. This novel model uses Harris Hawk Optimization (HHO) to help select features and a Recurrent Fuzzy Neural Network (FRNN) for classification. The authors make a point to systematically check and compare the HHO-FRNN model with standard classifiers, like Support Vector Machine (SVM), Random Forest (RF), Long Short-Term Memory (LSTM), and standard Recurrent Neural Networks (RNN), on multiple well-known datasets such as IMDb and Twitter Sentiment140. Every model is evaluated by how well it deals with semantic ambiguity, variations in feature dimensions, and imbalanced number of classes. TF-IDF is first used to get feature vectors, which are then improved through HHO, while cosine distance is applied to help understand the sentiment context better. Researchers found that compared to other models, the HHO-FRNN model gathered higher accuracy, precision, recall, F1-score, and generalized across different messy and unstructured sources of text. While deep learning does well with lots of data, it is vulnerable to mistakes caused by noise and can memorize the details and overfit the training data. By comparison, the HHO-FRNN offers effective learning and can easily be understood, making it useful in harder sentiment contexts. The research finds that hybrid intelligent systems work better than traditional and deep approaches for real-world sentiment analysis.