Using Sentiment Analysis and Machine Learning to Collect the Perception of Online Learning

The aim of this research is to collect the perception of the Education community towards the effectiveness of online learning taking place during the period of Covid-19, confinement and online learning. This has been achieved through sentiment analysis complemented by machine learning techniques. Public opinion has been mined predominantly from the famous social media site Twitter. The data collected was analysed through a web based application that has been developed, so that the behaviour and perception of people around the world is known. The opinion that has been mined was then classified into three categories positive, neutral and negative. An analysis has been performed on the classified data to see what percentage of the population sample falls into each category and this can be visualised geographically. Existing Sentiment Analysis Systems were critically analysed before developing the Web application. Features such as tokenization, lemmatisation, removal of stop words, feature extraction and confusion matrix were used to perform sentiment analysis. During this research, it was observed that features like removal of slang words, abbreviation, POS tagging and taking into consideration of negations are rarely present in existing Sentiment Analysis systems and these features have now been successfully implemented. These features greatly help to increase the accuracy of the supervised machine learning models. The results obtained were encouraging and it was observed that Random Forest classifier had better recall, precision and accuracy as compared to other techniques such as Bayes, Linear Support Vector Machine and Decision Tree.