<![CDATA[The growing popularity of social commerce platforms such as TikTok Shop Live offers an interactive shopping experience through live streaming and chat features. However, many messages in the chat are irrelevant to sales, often containing spam or meaningless comments. This study examines the effectiveness of Natural Language Processing (NLP) methods in distinguishing between sales-related and spam chats to improve seller–consumer interactions. A systematic literature review was conducted on 13 articles focusing on spam detection and chat classification. The findings reveal that classical methods such as Naïve Bayes with TF-IDF achieve an accuracy of 85–90%, while K-Nearest Neighbor (KNN) and Logistic Regression are effective for simple cases with an accuracy of 80–87%. Deep learning methods deliver higher performance, with Long Short-Term Memory (LSTM) achieving 90–93% due to its strength in recognizing sequential patterns and informal language, while Convolutional Neural Networks (CNN) reach competitive accuracy rates of 88–91%. Transformer-based models, particularly Multilingual BERT, yield the highest accuracy (93–95%) because of their ability to capture contextual meanings in informal Indonesian texts. These findings confirm that a combination of classical and modern NLP methods is effective in supporting automated TikTok Live chat detection systems and can be further developed into real-time applications to enhance interaction quality, data-driven decision-making, and consumer satisfaction.]]>
