Aspect based Sentiment Analysis using Attention based BiGRU with Polarity Classification on Optimized Features

Abstract

Sentiment analysis is crucial for understanding public opinion, especially in the context of e-commerce and the growth of online enterprises. Initial methodologies saw sentiment analysis as a classification challenge at the document or phrase level, which are unable to encapsulate complex sentiments regarding individual elements. This constraint was mitigated by the advancement of Aspect-Based Sentiment Analysis (ABSA), which associates sentiment with particular aspects that are either expressly or implicitly referenced in the review. Nonetheless, substantial characteristics impede and diminish the sensitivity of analytical mood, rendering the selection and classification of features challenging. Consequently, feature selection is crucial in sentiment analysis. This study introduced the attention-based Bidirectional Gated Recurrent Unit (BiGRU) for sentiment analysis to determine polarity of the comment. The research utilizes the SenEval2016, MAMS, YASO, and DOTSA datasets. Initially, pre-processing is conducted by tokenization, part-of-speech tagging, and lemmatization. Subsequently, Bag of Words (BoW), Term Frequency-Inverse Document Frequency (TF-IDF), and Non-negative Matrix Factorization (NMF) are utilized for feature extraction. Subsequently, feature selection is conducted with the Interactive Autodidactic School Optimization Algorithm (IASOA). Ultimately, in the classification phase, an attention-based BiGRU is introduced for categorizing the selected features into distinct polarity classes. The results clearly indicate that the suggested work has surpassed previous approaches, achieving average accuracies of 97.8% and F-measure rates of 96%.