Sentiment Analysis in Social Media using Bidirectional Long Short Term Memory with Focus Loss function

Sentiment Analysis (SA), is often known as "Opinion Mining" which seeks to analyze sentiment polarities in opinion-based texts. Several Deep Learning (DL) approaches are developed to effectively solves the complex SA challenges. However, existing methods failed to identify the sentiments correctly due to contextual embedding’s and different meaning for same word in a sentence or document. To overcome this, a Bidirectional – Long Short Term Memory with Focal Loss (Bi-LSTM-FL) function for SA in social media. The proposed Bi-LSTM-FL model enhanced the SA process by capturing the context from both past and future words in a sentence which lead to understand complex sentence structures. The twitter dataset is utilized and preprocessed by to enhance the text format for classification. The preprocessed texts are fed to Principal Component Analysis (PCA) for selecting optimal set of features and to reduce dimensionality of features. Finally, with the help of selected features the tweets are categorized as positive, negative or neutral effectively. The experimental results of proposed Bi-LSTM-FL achieved accuracy of 98.2% which is greater when compared to previous method such as Gated Attention Recurrent Network (GARN).