Comparison of Text Representation Methods for Sentiment Analysis Using Support Vector Machine

This study aims to analyse the sentiment of text from hashtags on TikTok regarding public services in Lampung Province, categorised into three groups: positive, negative, and neutral. Data is obtained from comments on TikTok. TikTok is a social media platform that offers users unique and engaging special effects. Recently, netizens were stirred by a viral TikTok video criticising Lampung's poor road conditions, titled 'Alasan Lampung Tidak Maju-maju' (Reasons Lampung is Not Progressing). This video sparked a range of comments from netizens, including supportive, critical, and neutral responses. The study employs the KDD (Knowledge Discovery in Database) method to extract insights from the existing database. The collected data will be manually labelled using the Support Vector Machine algorithm and Python programming software before being classified. The findings show that the classification model's accuracy differs based on the text representation technique. Of the three word-to-vector techniques, the Bag of Words method reached 48% accuracy, TF-IDF achieved 71%, and FastText achieved 50%. In summary, the sentiment classification model for public service content in Lampung Province on TikTok reveals that the Support Vector Machine combined with the TF-IDF method delivers the highest accuracy.