<![CDATA[Indonesia’s gaming industry is rapidly expanding and produces extensive textual data from diverse sources such as news articles and social media. Named Entity Recognition (NER) models can extract valuable information from this data; however, general-purpose models remain suboptimal for the gaming domain due to its unique terminology. This study evaluates the impact of domain shift on the NERGrit model, a standard NER model from the IndoNLU benchmark, when applied to an Indonesian gaming text corpus. The model was tested on the gaming-domain corpus and compared with a domain-specific lexicon to identify error patterns through qualitative and quantitative analyses. Results show that although NERGrit can detect numerous entities, it often fails to classify them correctly. The dominance of the MISC category (61.8%) and recurring issues such as misclassification, entity boundary errors, and ambiguity between fictional and real entities indicate the model’s limitations. This study confirms the existence of domain adaptation challenges and introduces a new entity schema covering the categories GAME, PLATFORM, TECH, EVENT, CHAR,and COMPANY. The proposed schema provides a foundation for developing a more relevant NER dataset and model tailored to Indonesia’s gaming industry ecosystem.]]>
Copyrights © 2025
