Mohd Asad . P. Radha Krishna

Mohd Asad Department of Computer Science and Engineering, National Institute of Technology Warangal, Telangana, India E-mail: asadmohd1995@gmail.com, P. Radha Krishna Department of Computer Science and Engineering, National Institute of Technology Warangal, Telangana, India E-mail: prkrishna@nitw.ac.in

Received in final form on December 15, 2022

Abstract
Emotions represent the physiological states of humans due to the occurrence of an event. Emotions can be expressed in several ways, such as text, speech, facial expression, or a combination of the above types. In this work, we handled emotion detection in text conversations. This is a challenging task owing to text ambiguity, the presence of intertwined emotions, context dependence, etc. We present an adaptive context-driven emotion detection by employing Bidirectional GRUs and applied to Microsoft’s Dyadic conversation multi-class classification dataset. In our approach, we employed three Bi-GRUs to concatenate the embeddings obtained from Word2Vec (300-dimensional), Fasttext, and 200-dimensional fine-tuned and fed into a fully connected neural network. The experimental analysis demonstrated that our proposed approach with distant pre-training achieved the best mean F1-score of 76.5% and outperformed the remaining models.

Keywords
Long Short-Term Memory (LSTM), Word2vec, Gated Recurrent Unit (GRU), Convolutional Neural Network, Recurrent Neural Network (RNN).

Cite This Article
Mohd Asad and P. Radha Krishna, An Adaptive Context Driven Emotion Detection in Conversations Using Bidirectional GRUs, J. Innovation Sciences and Sustainable Technologies, 3(1)(2023), 1 - 13. https://doie.org/10.0421/JISST.2023516798

    160    40    Download