| Abstrakt: |
Twitter boasts 319 million daily active users, making it an invaluable asset for public figures and businesses looking to cultivate positive public image. Businesses can leverage sentiment analysis for real-time polling on various social media platforms allowing them to gauge public sentiment and opinion accurately. Recently, academic researchers focused on sentiment analysis as an approach for Twitter propaganda analysis. Text sentiment analysis is an automated process that offers valuable insight into the content of a text segment. It can reveal whether it conveys factual or subjective information and reveal its polarity; for Twitter sentiment classification this goal primarily lies with determining whether tweets have positive or negative undertones; researchers utilize various Machine Learning (ML), Deep Learning (DL) and other models to accomplish this task. Present research work utilising classification algorithms such as Support Vector Machines are among the most frequently utilized ML/DL models for sentiment analysis, along with Random Forest, Ensemble Machine Learning, Artificial Neural Networks, Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM) for effective classification. Also, for preprocessing the tweets API, techniques such as filtering, tokenization, removal of stopwords, stemming and lemmatization have been used. Then preprocessed input is fed as input to the TF-IDF and Bag of Words for vectorize the input. Then classification has been performed with aforementioned models. Finally, performance evaluation metrices have been perfomed, from that out of all these models used for sentiment analysis on Twitter dataset, Bidirectional LSTM has proven itself most accurate at detecting sentiment with an accuracy rate of 98.14%, 98.39% in vectorize techniques includes TF-IDF and Bag of Words—making this tool invaluable when conducting voice analyses on this platform. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink