Twitter sentiment analysis using aspect-based bidirectional gated recurrent unit with self-attention mechanism

Abstract

Nowadays, social media sites and applications such as Twitter, YouTube, Facebook and blogs gained much attention, because organizations use this huge information to monitor the user opinions. Therefore, the researchers focused on sentiment analysis to help organizations for better production. But, the existing studies concentrated only on document-level in the analysis, where the details of the sentiments are not considered. In order to perform a finegrained analysis, this research study uses the aspect-based sentiment analysis on Twitter data for finding the opinions of users. In general, a raw tweet contains stop words, Uniform Resource Locator (URL), emoji that are reduced in the stage of pre-processing. The polarities of pre-processed tweets are identified, then two important feature extraction techniques are used to extract the useful information. The specific sentiments with various aspects of services are extracted by using aspect-based feature extraction, which are given as an input for Bi-directional Gated Recurrent Unit (BGRU) for classification. The learning rate of BGRU is further improved by incorporating the self-attention layers. To test the efficiency of the proposed method, the experiments are conducted on real-time collected tweets in terms of accuracy, precision, recall and f-score. The results showed that the developed method achieved 80.4% accuracy, 81% precision, 80% of recall and f-score for 1000 dataset length when compared with traditional techniques: Long-short term Memory (LSTM) and standard GRU.