Humor Detection Using a Bidirectional Encoder Representations from Transformers (BERT) based Neural Ensemble Model

A lot of research has been done to aim to find out what makes someone laugh in a text. In recent years, detecting humor in written sentences has shown to be a fascinating and challenging endeavor. We describe a mechanism for identifying humor in brief texts in this paper. We employ a Bidirectional Encoder Representations from Transformers (BERT) architecture because of its benefits in learning from sentence context. Our proposed methodology also uses some other embedding models e.g., Word2Vec or FastText to generate Embeddings for sentences of a given text and uses these Embeddings as inputs in a neural ensemble network. We illustrate the efficacy of this methodology. We significantly reduced our root mean squared error by using this technique.