Seybold Report ISSN: 1533-9211
1M. Subathra, 2Dr.J. Vandarkuzhali, 3Dr. K. Meenakshisundaram
1Ph.D Research Scholar, 2Assistant Professor, 3Associate Professor and Head,
Department of Computer Science, Erode Arts and Science College (Autonomous), Erode, India
1iamsubathra@gmail.com, 2kuzhalijv@gmail.com, 3lecturerkms@yahoo.com
Vol 17, No 09 ( 2022 ) | DOI: 10.5281/zenodo.7074332 | Licensing: CC 4.0 | Pg no: 944-954 | Published on: 13-09-2022
Abstract
Sentimental Analysis is growing rapidly across various domains as a direct outcome of natural language processing and machine learning approaches for assessing and classifying emotions from data. Analyzing sentiment over product reviews is the clumsy task, which provides a variety of information regarding customer lifestyle. Hence high level of impedance involved while classifying the sentiment from the social media reviews documents. Existing researches have been focused on summarization of text, reduction of features, and prediction of sentiment separately. In this research work, all the approaches are integrated to provide a novel sentimental analysis framework for classifying reviews of a customer. The proposed work is consisting of three folds. Initially, pre-processing is done which includes tokenization, stemming, lemmatization, stop words, lower case conversion. Second, Selection of features is done with help of Effective Moth Flame Optimization Algorithm (EMFO), hence as a result standard features only exist which determines fittest individuals. Finally, the accurate sentiment prediction is done with help of Enhanced Recurrent Neural Network (ERNN). The proposed Heed BiGRU incorporate with BiGRU Layer and Attention layer. The proposed work is experimented by utilising the datasets, which is widely consisting of product reviews from social media platforms. To analyse the proposed customer sentimental analysis method, the evaluation metrics like accuracy, precision, recall and f-score are used and, also comparison has been made with existing state-of-art model. Proposed work outperforms well than other methods in terms of all performance metrics.
Keywords:
Sentimental analysis, Deep learning, Bidirectional Gated Recurrent Units, Heed Mechanism, Effective Moth flame Optimization.