Optimizing Sentiment Analysis in Hindi Poetry: A Hybrid Model Unifying Deep Learning, Machine Learning, and Metaheuristic Techniques

• Author(s): Vinod kumar ; Archismita Ghosh ; Kandikattu Sai Rachana ; Teetas Bhutiya ; Sukanya Wattal
• Paper ID: 1705305
• Page: 193-207
• Published Date: 16-12-2023
• Published In: Iconic Research And Engineering Journals
• Publisher: IRE Journals
• e-ISSN: 2456-8880
• Volume/Issue: Volume 7 Issue 6 December-2023

Abstract

Sentiment analysis, an automated computational methodology employed for the investigation and assessment of sentiments, emotions, and feelings conveyed in comments, feedback, or critiques, utilizes machine learning techniques to discern text patterns proficiently. This research leverages supervised machine learning, specifically exploring its application in the sentiment analysis of Hindi poetry-based text through the validation of model feasibility and accuracy using the Hindi Poetry Sentiment Corpus. The study delves into the examination of prevalent supervised machine learning techniques, including Multinomial Naive Bayes, Logistic Regression, and Random Forest, alongside deep learning methodologies such as Long Short-Term Memory and Convolutional Neural Networks. To evaluate classifier performance comprehensively, standard datasets are utilized, and metrics such as precision, recall, F1-score, RoC curve, accuracy, running time, and k-fold cross-validation are employed. This analytical approach yields valuable insights into the efficacy of diverse deep learning techniques, aiding practitioners in selecting suitable methods tailored to their specific applications. Furthermore, the investigation incorporates the application of the metaheuristic-based Grey Wolf Optimization technique to discern optimal features from pre-processed data. The genesis of "deep learning" (DL) in artificial neural network research is acknowledged, wherein word vectors trained by Word2Vec are utilized for the input layer (IL) and input into the CNN-LSTM joint model. Subsequently, the output of the joint model undergoes weighting and summation through self-attention before entering the SoftMax classifier, facilitating the emotion classification of the text. Rigorous comparative experiments validate the utility of the proposed model, demonstrating its superior performance over three comparison models [CNN, LSTM, CNN-LSTM] across various evaluation indices. Comparisons with other machine learning techniques, including Random Forest, Logistic Regression, Naive Bayes, CNN, and LSTM, reveal notable accuracies. Specifically, Random Forest, Naive Bayes, CNN, and LSTM achieve accuracies of 87.75%, 85.54%, 91.46%, and 88.72%, respectively. Notably, the proposed ensemble hybrid model attains the highest classification accuracy of 95.54%, precision of 91.44%, recall of 89.63%, and F-score of 90.87%, showcasing its efficacy in sentiment analysis applications.

Keywords

Hindi poetry-based text sentiment analysis, Machine Learning, Deep Learning, Grey Wolf Optimization, natural language processing, CNN-LSTM multi-feature fusion.

Citations

IRE Journals:
Vinod kumar , Archismita Ghosh , Kandikattu Sai Rachana , Teetas Bhutiya , Sukanya Wattal "Optimizing Sentiment Analysis in Hindi Poetry: A Hybrid Model Unifying Deep Learning, Machine Learning, and Metaheuristic Techniques" Iconic Research And Engineering Journals Volume 7 Issue 6 2023 Page 193-207

IEEE:
Vinod kumar , Archismita Ghosh , Kandikattu Sai Rachana , Teetas Bhutiya , Sukanya Wattal "Optimizing Sentiment Analysis in Hindi Poetry: A Hybrid Model Unifying Deep Learning, Machine Learning, and Metaheuristic Techniques" Iconic Research And Engineering Journals, 7(6)