Sentiment analysis of emoji fused reviews using machine learning and Bert

Abstract

The usage of Natural Language Processing (NLP) technology powered by Artificial Intelligence in processing of customer feedback has helped in making critical decisions for business growth in the aviation sector. It is observed that in many of the cases, emojis and emoticons are found to convey a lot of significant information about the user's opinion or experience regarding a certain product, a service or an event. Consequently, it is very much essential that these emojis/emoticons are considered for processing because they are found to play a vital role in sentiment expression, often conveying more explicit information than the text alone. Their inclusion helps in capturing nuanced sentiments, improving the overall accuracy of sentiment classification. In Spite of the fact that these elements are a significant part of the review comment provided by the customer, it is a common practice among the contemporary researchers to eliminate them right at the data-cleaning or the preprocessing stage. With an objective to provide a solution to the above drawback, we present a novel approach that performs sentiment analysis, with effective utilization of emojis and emoticons, upon the US Airline tweet dataset using various Machine Learning classifiers and the BERT model. Finally, the proposed model was evaluated using various performance metrics and achieved 92% accuracy, outperforming contemporary state-of-the-art frameworks by 9%.

Keywords: BERT; Emojis; Machine learning; Sentiment analysis; TF-IDF.

Fig. 1

Sample tweets and their translation using in-built and customized emoji dictionary EMOJIXT.

Fig. 2

Block Diagram of the proposed work.

Fig. 3

(a) CBOW Model. (b) Skip-gram Model.

Fig. 4

Architecture of BERT-base Model.

Fig. 5

Working Principle of Variant I, II and III.

Fig. 6

(a) Distribution of dataset into different classes. (b) Dataset instances with /without emoji.

Fig. 7

(a) Accuracy comparison of BERT-base Model on Variant II (b) Loss comparison of BERT-base Model on Variant II.

Fig. 8

(a) Accuracy comparison of BERT-base Model on Variant III (b) Loss comparison of BERT-base Model on Variant III.

Fig. 9

(a) Training accuracy of BERT-base model on Variant II vs. Variant III (b) Validation accuracy of BERT-base model on Variant II vs. Variant III.

Fig. 10

(a) Training loss of BERT-base model on Variant II vs. Variant III. (b) Validation loss of BERT-base model on Variant II vs. Variant III.