| dc.description.abstract | Skincare has been an important issue for women for a long time. The large number of local products makes consumers more selective in choosing the products they use. Product reviews are a reference when buying skincare. The reviews taken are from Twitter. This research aims to determine the sentiment analysis of local skincare reviews on Twitter using the Bert approach. This research also addresses the problem of negated words in user reviews so that the results are better with the method used, namely by marking words that contain negation. This research uses the Bert Approach, BERT is a pretrained language model, an architecture, and can also be called a state of the art method in the world of natural language processing and is a model that can study text contextually. The population is tweets originating from Twitter with sampling using tweet data containing reviews of local MS GLOW skincare of 1389 MS Glow skincare review data. With test results with three epochs 5, 10, and 16. Epoch 16 produces the best results, so epoch 16 is used to analyze sentiment. Test results using the Bidirectional Encoder Representations from Transfromer (BERT) method produced 80% accuracy using hyperparamters, namely batch size of 16, learning rate 5e-6, and epoch 16. The best accuracy results were obtained by epoch 16, namely 80% | en_US |
