UCLA

With the rapid advancement of technologies in the recent few decades, more and more students are entering the Science, Technology, Engineering, and Mathematics (STEM) field in college. As one of the leading universities in the world, University of California, Los Angeles (UCLA) has a strong group of undergraduate programs in STEM. With the size of the STEM community expanding rapidly, it is important that we examine the sentiment of the community through some statistical analyses. Among various approaches, sentiment analysis of the course reviews could help us understand the feedback from the student community, which could also provide us with many interesting insights. In this paper, we will look at the methodologies and results from applying the sentiment analysis pipeline on a corpus with around 7000 course reviews collected from the UCLA Bruinwalk website, with the goal of analyzing if students were happy (positive sentiment) or unhappy (negative sentiment) towards the STEM courses they have taken. As we shall see, because the reviews obtained from the Bruinwalk website were not initially annotated, the sentiment analysis pipeline consisted of two main components, review annotation and sentiment analysis modeling. Throughout the study, various data visualization techniques were utilized to help us obtain a better understanding of structure of corpus, including the features and the sentiment annotation. Multiple NLP model architectures, such as CNN+LSTM, Transformer, and the state-of-the-art BERT and DistilBERT architecture, were established and compared to optimize the sentiment prediction performance. The results from the sentiment analysis showed that around 60\% of the reviews collected contained a positive sentiment, the sentiment of the reviews was positively associated with the student grades, together with several other interesting findings.