K-means Clustering of Tweet Emotions: A 2D PCA Visualization Approach

Abstract

With the growing prominence of social media as a platform for expressing opinions and emotions, understanding the emotional undercurrents in large volumes of text data has become increasingly crucial. Tweets, often reflecting public sentiment, contain a rich tapestry of emotions that can be harnessed for diverse applications ranging from market analysis to mental health monitoring. The dataset comprises 40,000 tweet records, each tagged with one of thirteen distinct emotions, making it a challenging task to perform multiclass emotion classification due to the sheer volume of data and the nuanced spectrum of emotional expressions. Traditional classification models often struggle with such high-dimensional, multi-category data, leading to the need for a more sophisticated approach that ensures both accuracy and computational efficiency. To address the complexity of multiclass emotion classification, we propose a novel approach that combines text preprocessing, advanced feature extraction using TF-IDF, and dimensionality reduction via 2D Principal Component Analysis (PCA). We then apply K-means clustering to the reduced feature set to identify inherent groupings within the emotional content of the tweets. This method not only reduces computational demands but also logically consolidates the emotions into fewer categories, potentially enhancing the performance of subsequent classification models. The implementation of our method yielded distinct clusters that suggest a logical grouping of the emotions within the tweets. The 2D PCA visualization revealed clear separations among clusters, indicating that our approach successfully captured meaningful patterns in the dataset. The ability to effectively cluster complex emotional data opens the door to creating more nuanced and efficient multiclass classification models. By reducing the number of categories and focusing on clustered groups, we can streamline the classification process and enhance the interpretability of results. This has significant implications for real-world applications, including targeted marketing campaigns, public policy development, and mental health assessment from social media content.