A framework for sentiment analysis of textual data and emoticons on social networks

Document Type : Original Article

Authors

1 K.N.Toosi University of Technology

2 , K.N.Toosi University of Technology, Tehran, Iran,

10.22034/jasp.2023.53172.1201

Abstract

Nowadays, people use emoticons in the text to increase their feelings or summarize expressions. Earlier machine learning techniques only involve the classification of text, emoticons, or images solely, whereas emoticons with text have most of the time been neglected. Research shows deep learning was rarely applied in sentiment analysis on text and emoticon data combination. Therefore, this article has analyzed the text and emoticons separately and in combination to find the emotions. First a comparative analysis of convolutional neural networks and long-short term memory networks is performed, to increase the accuracy of word embedding, pre-trained word embedding vectors word2vec, glove, BERT and RoBERTa were used. Then, a new hybrid attention-based model is presented, which uses convolutional neural networks and Bidirectional long-short term memory networks. The results show that considering emoticons as a feature increases the accuracy of the model, also the proposed model has performed better than the basic models and previously reviewed works based on the four evaluation criteria of accuracy, precision, recall, and f1 score that accuracy of the proposed model on the dataset without emoticons is 85.80 and on the dataset including emoticons is 90.18.

Keywords

Advanced Signal Processing

Articles in Press, Accepted Manuscript
Available Online from 12 July 2023

Files

Share

How to cite

Statistics