<![CDATA[The high volume of unstructured conversations regarding adolescent mental health on platform X complicates manual topic identification. This study aims to map these public discourses using Latent Dirichlet Allocation (LDA) within the CRISP-DM framework. The corpus comprises 8,792 Indonesian tweets (combining Apify and Kaggle datasets) processed through noise removal, slang normalization, and bigram detection. Combined Coherence Score evaluation and qualitative interpretation identified an 11-topic model ( = 0.4880) as the optimal architecture. This validity was further corroborated by two independent raters against an alternative model (K=15), confirming the 11-topic model achieved comparable or superior coherence. The most dominant topics were general restlessness with self-soothing efforts (35.26%) and clinical anxiety with daily emotional pressure (18.16%). Furthermore, the study revealed that fandom-related terms are distributed across multiple topics, indicating that online community engagement functions as a cross-cutting element within adolescent mental health discourse.]]>
Copyrights © 2026
