Research Area:  Machine Learning

Abstract:

Sentiment analysis is one of the most challenging tasks in natural language processing (NLP). The extensively used application of sentiment analysis is sentiment classification of reviews. The purpose of sentiment classification is to determine the sentiment polarity of user opinion, attitude, and emotions expressed in the form of text into positive, negative and neutral polarities. Many advanced deep learning approaches have been proposed to solve sentiment analysis problem. Recurrent neural network (RNN) is one of the popular deep learning architectures which is widely employed in sentiment analysis. In this paper, we proposed a Two State GRU (TS-GRU) based on feature attention mechanism that concentrates on identifying and categorization of the sentiment polarity using sequential modeling and word-feature seizing. The proposed approach integrates pre-feature attention in TS-GRU to associate the complex connection between words by sentence based sequential modeling and capturing the keywords using attention layer for sentiment polarity. Subsequently, a decoder function has been added in the post-feature attention GRU, in order to extract the predicted features during attention mechanism. The proposed approach has been evaluated on three benchmark datasets including IMDB, MR, and SST2. Experimental results conclude that the proposed TS-GRU model obtained higher sentiment analysis accuracy of 90.85%, 80.72%, and 86.51% on IMDB, MR, and SST2 datasets, respectively.

Keywords:  

Author(s) Name:  Zulquarnin,Muhammad,Ghazali,Rozaida,Aamir,Muhammad Azim,Yana Mazwin Mohammad

Publisher name:  EBSCO

DOI:  10.1007/s11042-022-13339-4

Volume Information:  volume 83

Paper Link:   https://openurl.ebsco.com/EPDB%3Agcd%3A5%3A5909813/detailv2?sid=ebsco%3Aplink%3Acrawler&id=ebsco%3Adoi%3A10.1007%2Fs11042-022-13339-4