HARC-New Hybrid Method with Hierarchical Attention Based Bidirectional Recurrent Neural Network with Dilated Convolutional Neural Network to Recognize Multilabel Emotions from Text

Abstract

We present a modern hybrid paradigm for managing tacit semantic awareness and qualitative meaning in short texts. The main goals of this proposed technique are to use deep learning approaches to identify multilevel textual sentiment with far less time and more accurate and simple network structure training for better performance. In this analysis, the proposed new hybrid deep learning HARC model architecture for the recognition of multilevel textual sentiment that combines hierarchical attention with Convolutional Neural Network (CNN), Bidirectional Gated Recurrent Unit (BiGRU), and Bidirectional Long Short-Term Memory (BiLSTM) outperforms other compared approaches. BiGRU and BiLSTM were used in this model to eliminate individual context functions and to adequately manage long-range features. Dilated CNN was used to replicate the retrieved feature by forwarding vector instances for better support in the hierarchical attention layer, and it was used to eliminate better text information using higher coupling correlations. Our method handles the most important features to recover the limitations of handling context and semantics sufficiently. On a variety of datasets, our proposed HARC algorithm solution outperformed traditional machine learning approaches as well as comparable deep learning models by a margin of 1%. The accuracy of the proposed HARC method was 82.50 percent IMDB, 98.00 percent for toxic data, 92.31 percent for Cornflower, and 94.60 percent for Emotion recognition data. Our method works better than other basic and CNN and RNN based hybrid models. In the future, we will work for more levels of text emotions from long and more complex text.