Implementasi Algoritma Topic Modeling pada Data Cuitan Terkait Kesehatan Mental pada Platform Media Sosial Twitter (X) Berbahasa Indonesia

This research aims to implement topic modeling algorithms on mental health text tweet data, reviewing the literature on this topic reveals that previous research has a narrow focus, causing many issues to be overlooked. This paper addresses these issues by applying topic modeling methods across various mental health frames through several stages. First, the paper examines various mental health frames resulting in 7 main labels: 'Awareness', 'Feelings and Problematization', 'Classification', 'Accessibility and Funding', 'Stigma', 'Service', and 'Youth'. The second stage focused on compiling a dataset of 29,068 Indonesian tweets, by filtering tweets using the keywords “mental health” and “mental health”. In the third stage, this research conducted data preprocessing and manual labeling of 3,828 randomly selected tweets, which were chosen due to the impracticality of labeling the entire data and so that each randomly drawn data has an equal chance of not referring to any one label. Finally, in the fourth stage, topic modeling experiments were conducted involving two scenarios using three different algorithms. Evaluation was done using human interpretation and coherence score. The results of the topic modeling experiments show that the LDA and LSI algorithms are more effective than HDP in generating keywords relevant to each main label. LDA shows an increase in coherence as the number of clusters increases up to a certain limit. LSI shows mixed coherence performance depending on the number of clusters, while HDP has relatively stable coherence from 3 to 7 clusters.