Sentiment Analysis for Brand Monitoring using Transformer with Bidirectional Long-Short Term Memory

In today’s digital era, social media platforms have become a crucial channel for customer to express their opinions and sentiments regarding the brands. Additionally, the sentiment analysis has developed as an important tool for monitoring brand reputation which enables companies to track online reputation to make informed decisions. However, traditional model struggled with the complexity and nuances of human language which raised as a significant challenge for sentiment analysis models. Therefore, Transformer with Bidirectional Long-Short Term Memory (Trans-BiLSTM) is proposed in this research. The transformer architecture effectively captures the contextual relationships while BiLSTM model captures sequential dependencies. The ability of both architectures accurately classifies the sentiments for brand monitoring. Initially, the input tweets are collected and then preprocessed to remove irrelevant information such as stop words removal, special characters removal. Consequently, the cleaned tweet is converted into tokens and fed into proposed Trans-BiLSTM for sentiment analysis. From the results, the proposed Trans-BiLSTM attained better results when compared to ensemble model in terms of accuracy (99.90%), sensitivity (99.15%), precision (99.80%) and F1-score (99.55%) respectively.