Sentiment analysis of TikTok reviews using lexicon- based methods with exploratory data analysis.

Abstract

This report takes a closer look at sentiment analysis using a lexicon-based approach combined with exploratory data analysis (EDA) to make sense of text data. The goal is to figure out the patterns and trends in user reviews, breaking them down into positive, negative, and neutral sentiments. For the analysis, Textblob were used to assign polarity scores to the text, helping to understand the emotions and opinions expressed. The EDA methods added an extra layer by uncovering word patterns, frequency trends, and sentiment changes over time. Visual tools like word clouds, bar charts, time series analysis and violin plots made it easier to spot key insights. One significant finding is that shorter reviews were mostly positive or neutral, while longer reviews often pointed out negative feedback or detailed issues. However, the analysis also highlighted some challenges, like handling sarcasm, complex language, and context, which could sometimes lead to misinterpretation of sentiments. To deal with these challenges, the report suggests adding word embedding models and machine learning algorithms to improve accuracy and context understanding. It also recommends using sarcasm detection models to better handle ambiguous comments. Overall, this study shows that combining lexicon-based sentiment analysis with EDA techniques can provide valuable insights into text data. It also highlights areas for future improvements, such as hybrid methods, custom lexicons, and advanced NLP techniques to make sentiment analysis even more accurate, scalable, and robust.