A Study on Deep Learning based Classification and Identification of offensive memes

Sarcasm is a complicated language device that is regularly utilised on e-commerce and social media websites. Failure to recognise sarcastic statements would confuse classification algorithms and result in incorrect findings in Natural Language Processing applications like sentiment analysis and opinion mining. Studies on sarcasm detection have included a variety of learning techniques. However, the majority of these teaching strategies have always focused their attention entirely on the concepts that were communicated, ignoring the context. As They thereby missed the meaning and context of the sarcastic statement. Second, the emotional polarity of words is not taken into account when using the word embedding learning methodology, which is a common way for convolutional feature vector encoding in NLP deep learning approaches. To overcome the issues mentioned above,This research presents a context-based feature strategy for sarcasm detection that combines deep learning, BERT, and conventional machine learning. Two benchmark datasets from the Internet Argument Corpus (IAC-v2) and Twitter were used for the categorisation. Employed are all three learning models. Deep learning and embedding-based representation are both utilised by the initial model. Word embedding and context are being developed utilising global vector representation (GloVe) learning and Recurrent Neural Network (RNN) with a Bidirectional Long Short Term Memory (Bi-LSTM). A pre-trained Bidirectional Encoder representation (BERT) is used to build the second model, which is based on Transformer. In contrast, the third model is founded on the BERT feature’s feature fusion. a function that incorporates sentiment-related, syntactic, and GloVe embedding