Opinion mining of social media reviews using fuzzy inference and intent prediction using deep learning

Abstract

Sentiment Analysis (SA) or opinion mining has emerged with the quick elevation in internet users, and the increasing power of Social Medias (SM) and also online review sites. SA aids in finding what other people think and comment on. Recently, to capture the sentiment information on the feedback, numerous SA techniques are developed. The major downside of such techniques is lower accuracy and higher training time. This paper proposes a proficient SA technique in SM reviews to trounce these downsides. Originally, utilizing tokenization, lemmatization, Stop Word (SW) removal, together with URL removal, the input dataset is preprocessed. Then, from that preprocessed data, the emoticon and non-emoticon features are extracted. Subsequently, those resulting features are ranked grounded on their meaning. After that, the classification is done by utilizing Bell Shaped Gaussian Kernel Membership Function which is the Enhanced system of Adaptive Neuro Fuzzy Inference System and is called GKMFANFIS. Lastly, for identifying the user's reason behind the feedback, the Emotional Classification (EC) is performed utilizing the Modified Deep Learning Neural Networks (MDLNN). In MDLNN, Elephant Swarms Water Search Optimization (ESWSO) is applied to optimize the weight values. The proposed methodology achieves better performance and efficiency when weighed against prevailing methodologies regarding the accuracy, recall, precision, together with an F-measure. The accuracy of the proposed GKMFANFIS along with ANFIS is 92% and it shows better result when compared with the existing BERT system.