Hybrid Model for Topic Classification of English Learning Accounts on Instagram Using LDA and SVM

English is an important international language to master, especially when in places with different language speakers. However, according to the EF English Proficiency Index, English proficiency among Indonesians is still very low. Indonesia ranks 74 out of 100 countries. Nowadays, many people use social media platforms, like Instagram, to share information and learn. There are numerous English learning accounts on Instagram with thousands of posts, making it difficult to group the topics found in these posts. Therefore, topic modeling and classification need to be performed to determine which topics account owners should improve to attract learners' interest. In this paper, the authors used the Latent Dirichlet Allocation (LDA) method for topic modeling and Support Vector Machine (SVM) for classification. The authors also compared the results when the labeling was done by experts. The expert labeling resulted in 9 topics, while using LDA produced 6 topics. This indicates that the topics generated by the experts are more specific compared to LDA. However, the SVM classification accuracy results from the LDA model are higher compared to the manual labeling. Specifically, the SVM classification accuracy using the polynomial kernel is 79.33%, using the sigmoid kernel is 79.33%, and using the RBF kernel is 80.00% for the LDA model. On the other hand, the SVM classification accuracy using the polynomial kernel is 35.33%, using the sigmoid kernel is 33.67%, and using the RBF kernel is 33.67% for the manual labeling.