Multi-class Classification of Twitter Sentiments using Frequency based, LSTM and BERT Methods

Abstract

Abstract: Twitter is one of the most popular social media applications used by people of diverse age groups. Tweet has become an important mode of communication. Tweets are unstructured and often consist of different kinds of data tokens like mentions, hash tags, emoticons etc. Analysing the sentiment of this kind of unstructured data is challenging. In this paper, the experimentation of multiclass twitter sentiment analysis is presented, using three different approaches. Starting with the simple approach of frequency based feature extraction method, that helps capture the features from a text document by counting the frequency of the occurrence of words, to more advanced approaches like LSTM and BERT have been used. Different preprocessing techniques and different classifiers are explored in this paper. The performance of all the three methods is evaluated using metrics (precision, recall and f1 score). At the end the results of all three approaches are discussed and compared against one another. Keywords: NLP, BERT, LSTM, MLP Classifier, Multiclass Classification, Embeddings, Tweets.