Multi-Modal Emotion Extraction using Neural Network Hybrid Activation Function

Abstract

The higher availability of the online platforms for sharing feedback for various aspects or products, the demand for analyzing the contents to estimate customer satisfaction level is increasing. The service providers are keen to understand the product sentiment from these informal reviews, which as per them demonstrate the true sentiment for the product. Thus, a good number of research attempts can be seen for analyzing the texts for extracting the sentiments. Nonetheless, these methods ignorea few foundations facts by using standard methods for tokenization, lemmatization and further extraction of the sentiments using tagging methods and cause either underfitting or overfitting problems. Hence, this proposed method demonstrates a few unique strategies such as differential analysis for tokenization, reduced time complex lemmatization, threshold-based sentiment extraction and further summarization. Finally, this work outcomes into a 99% accuracy using enhanced sigmoid based neural network activations and unique strategy for weight correction in the neural networks.