Enhanced Mixed Emotions Prediction for Multilabel and Sarcasm Based Sentences using Deep Learning Techniques

Abstract

Sentiment analysis is a subdivision of natural language processing (NLP) that study newlinehuman emotions in written text. Unstructured text data is a primary challenge for newlineresearchers in identifying intense emotions. Typically, an abstract noun expresses an newlineintense emotion that human senses cannot feel and is found in complex sentences. newlineThis research focuses on obtaining intense (abstract noun) mixed emotions from social newlinemedia data and was conducted in three phases. Initially, in the first phase, to choose newlinethe best vectorization technique, as well as to identify the intense mixed emotions newlinesentences from the complex sentences, several vectorization feature selection models newlinewere developed using all possible combinations of Count, Term Frequency/Inverse newlineDocument Frequency (TF-IDF), Word to Vector (Word2Vec), Global Vectors (GloVe) newlinevectorization techniques along with the Support Vector Machine (SVM), newlineConvolutional Neural Networks (CNN) and Recurrent Neural Network (RNN) newlineclassifiers, which classified the three benchmark social media datasets (Sentiment 140, newlineIMDb, SST-5) into positive abstract noun complex sentences, mixed abstract noun newlinecomplex sentences, and mixed non-abstract noun complex sentences using four newlineproposed algorithms. It was found that GloVe vectorization technique outperforms the newlineother vectorization models. In the second phase, the proposed hybrid rule-based newlinemultilevel model was developed to classify the mixed complex sentences at twelve newlinelevels of intense mixed emotions based on the abstract noun and adjective sentences newlineusing the proposed adjective searching algorithm and deep learning classifiers [CNN, newlineRNN, Long Short-Term Memory (LSTM)] along with the pre-trained two word newlineembedding techniques [GloVe and Bidirectional Encoder Representations from newlineTransformers (BERT)]. In the third phase, to address the challenges in predicting newlinepolite sarcasm sentences, the proposed sarcasm - abstract noun based model was newlinedeveloped to classify the given dataset into the sarcasm- abstract noun and non_ newlinesarcasm-abstract noun int