Classifying claims from hindi and english tweet texts: ml and dl tricks

Abstract

This thesis, ”Classifying Claims from Hindi and English Tweet Texts: ML and DL Tricks,” tackles the challenge of classifying claim veracity in multilingual tweet texts using machine learning (ML) and deep learning (DL). We collect and pre-process a comprehensive dataset of Hindi and English tweets, annotated for veracity. Our research explores the performance of ML and DL models, including traditional classifiers and neural networks, in differentiating true and false claims. Additionally, we delve into advanced techniques like transfer learning and ensemble to enhance model accuracy. The results showcase the potential of these ”tricks” to improve classification across languages. This study contributes to the vital field of social media information verification and offers insights into the development of effective tools for combating misinformation on platforms like Twitter. This thesis highlights the significance of harnessing both ML and DL techniques, along with the employment of ”tricks” to enhance their performance, in addressing the pressing issue of claim classification in multilingual tweet texts. Apart from this, the results on this database using a three classifiers and two models architecture. Three classifiers and two models are used: Logistic Regression, Decision Tree, SVM, LSTM, BERT. Among these three ML classifiers Decision Tree achieved 65% the best accuracy for English and two DL models BERT achieved the best accuracy for Hindi, i.e., 75% .