Social Media Trends Analysis using the Bi-LSTM with Multi-Head Attention

Abstract

In this global world, the usage of social media has produced a vast amount of human-generated data that will be analyzed to determine people’s sentiments. Sentiment analysis refers to the method of automatically grouping web data into different categories. The Proposed work presents bidirectional long-short memory (Bi-LSTM) network based on a multi-head attention mechanism to identify sentiments like business & economics, entertainment, science & technology, or health. We utilized a self-collected dataset from Twitter API. Bi-LSTM is used to capture two-way semantic information and the additional multi-head attention mechanism focuses on outputted information of Bi-LSTM. To assess the performance of the proposed work we utilized Precision, Recall, Accuracy, and f1-score as evaluation metrics. The proposed methodology is also contrasted with well-known sentiment analysis methods including Naive Bayes, Convolution Neural Network, Recurrent Neural Network, and LSTM our model performs best with 98.72% accuracy, 93.65% precision, 94.02% recall, and 93.20% f1-score.