Application of Fine-Tuned IndoBERT for Sentiment Classification Local Product Reviews on Tokopedia Marketplace with Limited Dataset

Abstrak

Abstract Analysis of sentiment to product reviews is one of the important approaches to evaluating consumer satisfaction with local products on e-commerce platforms. However, the limitations of Indonesian-speaking datasets are often a barrier to building classification models. This research aims to implement and test the performance of IndoBERT models that have gone through fine-tuning processes on limited datasets derived from local product reviews in Tokopedia. The fine-tuning process is performed using manually classified data sets into positive, negative, and neutral categories. Performance evaluation is performed by measuring accuracy, precision, recall and F1-score, as well as compared to baseline models such as Naïve Bayes and SVM. Research results show that re-trained IndoBERT is able to provide higher accuracy despite limited data conditions, signaling the effectiveness of transfer learning in the Indonesian domain. The findings contribute to the development of efficient and adaptive local language-based sentiment analysis system to data limitations. Kata Kunci : IndoBERT,Centiment classification,Tokopedia,Limited dataset,local products,fine-tuning,NLP

Identifier

• DOI: https://doi.org/10.59934/jaiea.v5i1.1629
• OpenAlex: https://openalex.org/W4415349550