KoBERT, KoGPT-2, KoBART 활용 및 하이퍼파라미터 최적화를 진행한 리뷰 감성분석 애플리케이션 구현 = Implementation of Review Sentiment Analysis Application Using KoBERT, KoGPT-2, and KoBART Optimized Hyperparameters

다국어 초록 (Multilingual Abstract)

User reviews and ratings available on application distribution platforms have a significant impact on the number of downloads an application receives, so developers rely on reviews to get feedback from users and update their applications. However, it is inconvenient to read all the reviews to know what users want. To improve this, we want to analyze the review dataset and show the results to developers. After cleaning the dataset, we proceeded to fine-tune the model by changing the hyperparameters. We created an initial dataset by crawling KakaoTalk and Instagram reviews, and conducted sentiment analysis using KoBERT, KoGPT-2, and KoBART models. We retrained each model with the purified dataset and changed the hyperparameters of the models to improve the learning. While the accuracy of sentiment analysis with the initial data was about 74%, we can see that the accuracy increased by about 15% to about 89% after data purification and model hyperparameter correction. We then developed an application to select and reference reviews using the model with the highest sentiment analysis performance. By using this application, we hope to help developers upgrade to improve user satisfaction.