Improving Sentiment Analysis in Digital Marketplaces through SVM Kernel Fine-Tuning

Abstract

The rapid growth of the online market, particularly in the digital realm, has spurred the need for in-depth studies regarding marketing strategies through public opinion, especially on platforms like Twitter. The sentiments expressed in customer tweets hold significant insights into their satisfaction or dissatisfaction levels with a service. Therefore, the use of ML algorithms in sentiment analysis is imperative to detect whether such comments lean towards positivity or negativity regarding a service. This research focuses on sentiment analysis towards three major e-commerce platforms in Indonesia: Tokopedia, Shopee, and Lazada, through the utilization of Twitter. The classification process involves various stages, including preprocessing, feature extraction and selection, data splitting for classification, and evaluation. The selection of both linear and non-linear SVM models as the focus of this research is based on their ability to handle large and complex datasets. The linear kernel is chosen for its proficiency in cases with a linear relationship between features and class labels, while the non-linear SVM provides flexibility in dealing with complex and non-linear relationships. Based on the evaluation results of the SVM model on the dataset, it is found that the polynomial kernel provides the highest accuracy value of 93%, with a training data share of 85%. This model features strong prediction capabilities with a precision of 93% for negative and 93% for positive labels. Although the linear kernel and other kernels showed solid performance, the polynomial kernel provided the most optimal results in the context of online marketplace sentiment analysis using data from Twitter.