Dominant lexicon based Bi-LSTM for emotion prediction on a text

Abstract

User-generated content and opinionative data has become a massive source of information on World Wide Web in the past few decades. Through social media people can share more conveniently their opinions, views, feelings and attitude about a product, person or event at anytime and anywhere as daily basis. This ever-growing subjective data makes enormous amount of unstructured data in web. Analyzing emotion in this raw unstructured data gives a very fruitful information for any kind of decision making process taken by both government and industries. Sentiment or emotion analysis is a field of Natural Language Processing (NLP), is used to identify the emotion depicted (by) in the form of text. Computation of emotion and emotion intensity depicted by a text is a very difficult task. Feature extraction from the text for vector representation is a difficult step of emotion analysis because it defines the emotion accuracy of the prediction. In this paper, a selective lexicon based BI-LSTM technique has been proposed. This technique uses only the most affected lexicon and its features for final vector representation. This method is a combination of features collected from the convolutional Neural Network (CNN), Long Short Term Memory (Conv-LSTM) and Bidirectional Long Short Term Memory (BI-LSTM). As a result the proposed model Selective Lexicon Based BI-LSTM (SL + BI-LSTM) outperforms all the models with high accuracy.