Machine Learning or Lexicon Based Sentiment Analysis Techniques on Social Media Posts

Social media provides an accessible and effective platform for individuals to offer thoughts and opinions across a wide range of interest areas. It also provides a great opportunity for researchers and businesses to understand and analyse a large volume of online data for decision-making purposes. Opinions on social media platforms, such as Twitter, can be very important for many industries due to the wide variety of topics and large volume of data available. However, extracting and analysing this data can prove to be very challenging due to its diversity and complexity. Recent methods of sentiment analysis of social media content rely on Natural Language Processing techniques on a fundamental sentiment lexicon, as well as machine learning oriented techniques. In this work, we evaluate representatives of different sentiment analysis methods, make recommendations and discuss advantages and disadvantages. Specifically we look into: 1) variation of VADER, a lexicon based method; 2) a machine learning neural network based method; and 3) a Sentiment Classifier using Word Sense Disambiguation, Maximum Entropy and Naive Bayes Classifiers. The results indicate that there is a significant correlation among all three sentiment analysis methods, which demonstrates their ability to accurately determine the sentiment of social media posts. Additionally, the modified version of VADER, a lexicon based method, is considered to be the most accurate and most appropriate method to use for the semantic analysis of social media posts, based on its strong correlation and low computational time. Obtained findings and recommendations can be valuable for researchers working on sentiment analysis techniques for large data sets.