Enriched Glove of Word Embedding and BERT of Contextual Embedding in Sarcasm Detection with AdaGrad Optimization Technique

The word Sarcasm related to other terms such as derision, irony, mockery, ridicule, scorn, contempt, bitterness, etc. is a challenging task in the area of Natural Language Processing (NLP). Even humans can find it difficult to understand the ironic words. The sentences may be categorized into positive sentiments. But in-depth contextual information points to negative or sarcasm. Lack of understanding of the contextual meaning of text and POS tagging and grammar may lead to critical tasks in identifying sarcasm. Nowadays, Sarcasm is a prominent crime, and being increases exponentially through social media in the form of text. Identifying the text as sarcasm is only by way of understanding the contextual information. After the preprocessing steps involved, features can be extracted through Word Embedding Techniques of GloVE(Global vectorization) and Contextual Embedding of the Transformer based approach, (Bi-Directional Encoder Representation of Transformer) BERT of neural networks models to classify the sentence as Sarcasm or Non-sarcasm. The main objective of the proposed work is to enrich contextual information with deep learning models like GloVe and BERT with the Adagrad optimization techniques. Furthermore, the work was tested with the SARC benchmark dataset and compared with related works which highlights the accuracy rate by increasing 7%.