Object-based content computing model for filtering communication elements indicated as Ethnic, Religious, Racial, and Intergroup (SARA) on social media

Social media posts in Indonesia vary widely in form and meaning, creating the potential for offensive expressions related to ethnicity, religion, race, and intergroup relations (SARAs), particularly when words considered neutral in one region are perceived as insulting in another. To address this challenge, this study develops a computational model that classifies and filters the SARA, which is related to offensive content via an object-oriented design, and a long short-term memory (LSTM)-based classifier, which is enhanced with fuzzy logic to incorporate the regional context into decision-making. The model processes a corpus representing linguistic variations across Indonesian regions and determines whether content should be suppressed on the basis of its classification and regional interpretation. The experimental results show strong model performance and indicate that 14.41% of the posts containing SARA elements were filtered by the system, demonstrating its potential for supporting content moderation in Indonesia’s multilingual and culturally diverse social media environment.