Modified Binary Particle Swarm Optimization Based Deep Hybrid Framework for Sentiment Analysis

Sentiment investigation or analysis constitutes a vital segment of the computerized processing of language that tries to capture public emotion from the available user feedback. This study suggests hybridized deep neurally connected network-reliant framework of sentiment assessment, in which we modify the Binary Particle Swarm Optimization and applied that Modified Binary Particle Swarm Optimization toward feature or attribute space optimization under the guidance of emotion knowledge retrieved from our specially designed SentiWordNet-based fitness estimator, following the preprocessing phase. This alteration makes it easier to circumvent the locally optimal choice and makes it viable to navigate to the globally optimal result with greater efficiency. Next, the textual features are mathematically represented by employing the pre-trained embedding technique to handle them effectively through Deep Neural Learning techniques. The encoded or embedded attributes or features are subsequently given to the deep hybridized neurally connected network. It consists of Convolutional Neural Network, which can create hierarchical depictions in order to capture locally embedded features and Long Short Term Memory, which seeks to look up relevant previous knowledge to carry out sentiment polarization. This hybridized integration makes it feasible to gain through both distinct neural aspects. The sentiment label is finally provided by this deep neural network. This proposed hybrid framework is assessed and analyzed with various avant-garde methods by different performance measures using multiple databanks to express its effectiveness.