Efficiency Determination of Various Machine Learning Techniques for Sentiment Analysis on Social Media Platforms

Abstract

The social networking sites have evolved beyond platforms for mere communication, having become prominent spaces where users share and express their thoughts, emotions, and opinions. This results in a vast volume of unstructured data, which holds significant potential for sentiment analysis in the emerging applications, such as business intelligence, healthcare, fashion forecasting, stock market prediction, and reputation management. A major challenge in these applications lies in effectively handling the unstructured text data, which often renders traditional statistical methods inadequate for meaningful analysis. Also, these methods typically fall short in performing robust text mining and accurate sentiment detection. In contrast, Machine Learning (ML) models, particularly Deep Learning (DL) architectures, are capable of learning hierarchical representations from data across multiple abstraction levels, leading to a more precise and efficient sentiment analysis. This paper presents a comprehensive review of the state-of-the-art ML-based sentiment analysis systems that utilize data from social networking platforms. The review focuses on several critical aspects, including the datasets employed, word embedding techniques adopted, DL models implemented, and performance metrics, such as F1-score, recall, precision, and accuracy.