Comparative Analysis of Naïve Bayes and K-NN Methods on Social Media Boycott Issue X Case Study: McDonald’s

Abstract

The boycott movement against McDonald’s, triggered by its alleged support for Israel during the conflict in Gaza, has generated significant public discourse, particularly on the social media platform X (formerly Twitter). This study investigates public sentiment regarding the boycott campaign by analyzing comments and reactions to related content. A total of 1,585 tweets were collected using techniques for web scraping and underwent a comprehensive pre-processing phase, encompassing cleaning, tokenization, filtering, and stemming. Sentiment categories, namely positive, neutral, and negative, are automatically assigned using a lexicon-based technique customized for the Indonesian language. Text data was transformed into numerical form through the Term Frequency-Inverse Document Frequency (TF-IDF) technique, followed by sentiment classification using two supervised machine learning algorithms: Naïve Bayes and K-Nearest Neighbor (K-NN). Evaluation of both models was conducted using a confusion matrix and classification metrics. The results show that the dataset is highly imbalanced, with 93.5% of the tweets labelled as negative, 6.1% as neutral, and only 0.3% as positive. The K-NN model achieved better performance than Naïve Bayes (NB), with an accuracy of 93%, a precision of 31%, a recall of 33%, and an F1-score of 32%. On the other hand, the Naïve Bayes algorithm reached 39% accuracy, 33% precision, 29% recall, and an F1-score of 22%. These findings highlight the dominance of negative sentiment toward McDonald’s and demonstrate the efficacy of the K-NN algorithm in sentiment classification in unbalanced datasets. The insights from this study can inform public relations strategies and corporate reputation management in the face of socio-political controversies.