Instagram is a social media that can be used to promote schools by sharing various documentation of school activities, but schools still have difficulty analyzing engagement to find out the audience's interests. This software development aims to identify topics from captions and analyze the like engagement of each topic. 3,900 caption data were collected from five school Instagram accounts in Palembang with Instaloader. The hLDA algorithm is implemented to identify topics from the caption data, and generate a new dataset that gives the topic information of each caption. This dataset was then classified using SVM and SVM-SMOTE. SMOTE is used to overcome class imbalance in order to improve classification results. In the classification process, the dataset is divided into 70% for training and 30% for testing, with evaluation based on F1-Score. The best results were obtained by SVM-SMOTE, with the best F1-Score value from hLDA 3 Level Dataset (13 labels), reaching 95.68% and the lowest value from hLDA 5 Level Dataset (8 labels), reaching 79.43%. Datasets that have more topics give better classification results. Based on the number of likes for each topic in the hLDA 3 Level Dataset, the most popular topic is topic 11, which includes school facilities, student uniforms, and entertainment events. This information can help schools further develop the most liked topics and improve the less liked topics.