<![CDATA[The identification of hate speech must be accompanied by the identification of social identity concepts. This study aims to provide an alternative corpus with text metadata and social identity based on relevant laws that are designed to be implemented in machine learning. Two key questions are addressed: what social identity semantic domains are realized in the corpus, and what are the accuracy measurement results from the corpus? To achieve these aims, the study adopts a mixed-methods approach: qualitative for the first question and quantitative for the second. This research falls under the broader umbrella of computational linguistics, utilizing semantic domain theory and natural language processing. The first approach shows that the corpus only contributes five out of nine formulated domains, dominated by negative (uncategorized), religion, and ethnicity. The second approach indicates suboptimal conditions in the annotation distribution of the corpus, despite an average accuracy rate of over 80%. This condition limits the model’s ability to generalize beyond the information within the corpus, especially regarding social identity categories that are not fully represented. This study differs from previous ones by focusing on data categorization based on more up-to-date legal sources. Future research could elaborate on this work by incorporating new language use concepts aligned with the corpus's original goal to detect hate speech.]]>
Copyrights © 2025
