A Hybrid Approach for Happy Emotion Analysis through Partition and Voronoi-density based Clustering and DistilBERT

In today’s day and age, prioritizing the emotion of "Happiness" has become imperative. The constant movement of life and the stress it brings, be it personal or professional, needs to be diluted with simple moments that are able to bring joy. This work presents an analysis of tweets sourced from the HappyDB database comprising 100,000 tweets. Based on varying degrees of happiness, these tweets are classified into 3 categories-Normal, Happy and Very Happy respectively. A hybrid approach of partition and voronoi density based clustering algorithm ECR-DBSCAN is employed to efficiently assign the respective classes or target labels. Fuzzy logic augmented partition approach gave a more flexible assignment of classes. The study involves fine-tuning of state-of-the-art Distil Bidirectional Encoder Representations from Transformers(DistilBERT) models for classification of tweets. The fine-tuned pre-trained DistilBERT model for tweet classification resulted in a notable validation accuracy of 87.1%.