SOCIOPHYSICS APPLICATION FOR SENTIMENT ANALYSIS OF FEMALE DAILY REGARDING TEA TREE OIL-BASED BEAUTY PRODUCTS USING MACHINE LEARNING

• Main Author: Rafelia Zahrah, Tsaniyah
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81441
• Institution: Institut Teknologi Bandung

Sociophysics models social phenomena using data-driven approaches, such as in the case of opinion dynamics. Female Daily is one of the social media used for sharing opinions and experiences related to beauty products. In the discussion forums provided by Female Daily, the topic of acne ranks second in popularity. Tea tree oil has been proven to have various antimicrobial activities to treat acne. This study conducts sentiment analysis of Female Daily regarding beauty products based on tea tree oil using Maximum Entropy (ME), Random Forest (RF), and KNearest Neighbors (KNN) methods. The analysis starts from web scraping, data preprocessing, TF-IDF, and data classification conducted in two stages. The first stage involves the accumulation of 10 datasets with varying numbers of training data. The MEmethod consistently achieves the highest accuracy compared to the other models. The highest accuracy value for ME is 0.7443, RF is 0.716, and KNN is 0.69. Next, the second stage is performed on each dataset by applying an 80% training data variation based on considerations of accuracy, precision, and potential overfitting. It is found that ME and RF models achieve the same accuracy value in some datasets. The highest accuracy for both models is observed with the Tea Tree 3-in-1 Wash Scrub Mask product, which is 0.75. Meanwhile, the accuracy of KNN remains below 0.7 overall, with its highest peak observed for Tea Tree Skin Clearing Body Wash and Tea Tree Flawless BB Cream products. Moreover, the number of positive sentiments is more than three times higher than negative sentiments, indicating that beauty products based on tea tree oil are well-received by the community.