Event Labeling Approach for Twitter Datasets Leveraging N-grams, Topics, and Machine Learning Algorithms for Enhanced Event Detection

Event detection on social media has emerged as a prominent research area with the potential to unlock valuable information spanning marketing tactics, event tracking, and analysis. However, the effectiveness of machine learning models heavily relies on accurately labeled data, necessitating researchers to undergo time-consuming data labeling processes. This research paper introduces a simple event labeling approach tailored for Twitter datasets, leveraging a combination of n-gram analysis, word frequencies, bilinear interpolations, data distribution visualization, and topic modeling. The algorithm is applied on the ‘Disaster on Social Media’ dataset covering range of disaster events and on events related to Cauvery river from India. The proposed algorithm demonstrates notable performance in labeling tweets. Validation of the results is carried out using five popular ML algorithms, namely Support Vector Machine (SVM), Random Forest, Naive Bayes, Bagging, and AdaBoost. The highest accuracy of 82% is achieved with SVM on the benchmark dataset ‘Disaster on Social Media’, highlighting its competency in event detection on multi-domain dataset. Moreover, SVM, Random Forest, and Bagging deliver remarkable accuracies of 97% on Cauvery dataset. These outcomes underscore the efficacy of the proposed methodology in event detection applications. This research contributes significantly to enhancing the comprehension of events expressed in tweets and their subsequent categorization, thus facilitating more efficient event prediction applications.