Sentiment analysis using various machine learning algorithms for disaster related tweets classification

Abstract

Once a crisis arises, people use social media platforms (such as Twitter) to communicate real-time updates. This data is incredibly helpful to disaster relief and response organisations and may offer rapid notifications for prioritising requests. Text mining and machine learning algorithms can scan enormous amounts of unstructured data created by social media outlets like Twitter to recognise disaster-related content based on keywords and phrases. One of the difficulties that algorithms may confront is determining whether the tweet content discusses actual disasters or uses these keywords as metaphors. As a result, this research aims to apply natural language processing (NLP) and classification models to discriminate between authentic and bogus disaster tweets. This dataset from the Kaggle website includes tweets about genuine disasters and fictional disasters. Four machine learning classifier methods were used: KNN, SVM, XGBoostand, and Naive Bayes. KNN offers the highest accuracy.