Analysis and recognition of textual emotion for Twitter dataset using transfer learning

Abstract

Textual expressions are frequently not only direct using emotional terms, but also derive from a grasp of the meaning of concepts and how they interact in the text document, making identifying an individual’s emotional state appear difficult but necessary several times. In order for organizations and individuals to be able to provide clients with optimal services, it is important for them to recognize the various emotions felt by individuals. There are variety of methods available to detect emotions in texts, from traditional machine learning models to deep neural networks. The previously described models were unable to capture the emotional connections between words. In this work we have proposed a noble model which captures semantic and emotion information using transfer learning technique. The proposed model used BERT technique for contextual information encoding and second is emo2vec which is used for emotion information encoding. The results reveal that the proposed model improves the system’s emotion recognition quality by a resulting in F1 score of 1.0.