An efficient way of text-based emotion analysis from social media using LRA-DNN

Abstract

Text devices are effectively and heavily used for interactions these days. Emotion extraction from the text has derived huge importance and is upcoming area of research in Natural Language Processing. Recognition of emotions from text has high practical utilities for quality improvement like in Human-Computer Interaction, recommendation systems, online education, data mining and so on. However, there are the issues of irrelevant feature extraction during emotion extraction from text. It causes mis-prediction of emotion. To overcome such challenges, this paper proposes a Leaky Relu activated Deep Neural Network (LRA-DNN). The proposed model comes under four categories, such as pre-processing, feature extraction, ranking and classification. The collected data from the dataset are pre-processed for data cleansing, appropriate features are extracted from the pre-processed data, relevant ranks are assigned for each extracted feature in the ranking phase and finally, the data are classified and accurate output is obtained from the classification phase. Publically available datasets are used in this research to compare the results obtained by the proposed LRA-DNN with the previous state-of-art algorithms. The outcomes indicated that the proposed LRA-DNN obtains the highest accuracy, sensitivity, and specificity at the rate of 94.77%, 92.23%, and 95.91% respectively which is promising compared to the existing ANN, DNN and CNN methods. It also efficiently reduces the mis-prediction and misclassification error.