Social media has emerged as a critical platform for understanding consumer behavior and purchase intentions. This study applies machine learning techniques, specifically text mining and sentiment analysis, to analyze social media data collected from Twitter, Instagram, and Facebook. Using a dataset of 10,000 entries, this research employs algorithms such as Random Forest, Naive Bayes, and Logistic Regression to classify consumer sentiments as positive, negative, or neutral. The results reveal that 58% of consumer sentiments are positive, which strongly correlates with increased purchase intentions, as indicated by a correlation coefficient (r) of 0.68. Negative sentiments show a significant adverse effect, with an r value of -0.45. The Random Forest algorithm demonstrated superior performance with an accuracy of 87%, outperforming other models. Additionally, the findings emphasize that sentiment trends differ across product categories, with health and beauty products receiving the highest proportion of positive sentiments (68%), while electronics faced notable negative feedback (28%). These insights offer practical applications for data-driven marketing strategies, enabling businesses to tailor campaigns and enhance consumer engagement. This study highlights the utility of machine learning for analyzing unstructured social media data and provides actionable recommendations for leveraging consumer sentiment in marketing strategies. Future research could expand to other platforms, such as TikTok or YouTube, and explore advanced deep learning models to improve predictive accuracy and incorporate multimedia data analysis.
Rahardjo et al. (Tue,) studied this question.