Twitter Data Analysis Using BERT and Graph-Based Convolution Neural Network

Twitter Data Analysis in Social media plays an essential role in spreading information during disasters and needy situations. May it be for seeking help or sharing and updating the seriousness of the situation or communicating the status or response to the public in need, social media plays a major role in case of emergencies. This research focuses on generating word embedding vectors using DistillBERT and generating a similarity matrix to construct the Graph-based Convolution Network model to classify text sequences and to analyse the performance of the DistillBERT with GCN model in Text classification. To implement this, contextual word embeddings are introduced for generating word vectors. The contextual word embeddings and the concept of graph neural networks has gained importance in capturing contextual relationship among the text to improve the performance of classification. This semi supervised model is used to detect and classify the need and availability of resource tweets with an accuracy of 96% as compared with state-of-the-art approaches.