Optimized RMDL with Transfer Learning for Sentiment Classification in the MapReduce Framework

Abstract

A core task in sentiment analysis is sentiment categorization, and it is crucial to understand user feelings based on their remarks in social media or product evaluations. Due to ambiguous phrases, refusal words, and other factors, categorizing sentiment presents several challenging issues. The objective of this research is to develop a hybrid optimization-based deep learning model and MapReduce framework-based sentiment categorization approach. The review document is taken from a dataset and used in this case with the MapReduce methodology. MapReduce is a software framework and programming model for analyzing massive volumes of data that consists of two phases, mapper and reducer. BERT tokenization and aspect term extraction are executed in the mapper phase, whereas sentiment analysis is performed in the reducer stage utilizing random multimodal deep learning (RMDL) with transfer learning and AlexNet and ResNet 50 as pre-trained models. In addition, the exponential coot political algorithm (ECPA) is offered as an optimization algorithm for weight optimization in RMDL. The ECPA is obtained by combining the exponential weighted moving average model (EWMA) with the coot algorithm, as well as a political optimizer (PO). The proposed ECPA_RMDL model has acquired 90.9% precision, 89.7% recall, and 89.9% f-measure.