PENERAPAN ZERO-SHOT LEARNING DENGAN DEEPSEEK-R1 DALAM ANALISIS SENTIMEN ULASAN DEEPSEEK AI DI GOOGLE PLAY STORE

This study aims to elevate the effectiveness of the Zero-Shot Learning (ZSL) approach using the DeepSeek-R1 model in performing sentiment classification on Indonesian-language reviews of the DeepSeek AI application from Google Play Store. Utilizing 2,000 unlabeled user reviews, the study employs instructional prompts to guide the model in categorizing opinions into three classes: positive, negative, and neutral. The model used is DeepSeek-R1-Distill-Qwen-1.5B, which is executed without fine-tuning. The results indicate that the model successfully classified 1,348 reviews with valid labels, with the majority being positive sentiment (83.9%). Performance evaluation showed an overall accuracy of 77.67%, with the highest F1-Score of 86.66% for the positive class. However, the model's performance declined on negative and neutral classes. This study demonstrates that DeepSeek-R1 holds potential for Indonesian sentiment classification via Zero-Shot Learning, although challenges remain in accurately identifying negative and neutral opinions.
Keywords : Zero-Shot Learning, DeepSeek-R1, Sentiment Analysis, Indonesian Language, Google Play Store