| 英文摘要 |
This is an era of advanced internet technology. Everyone has a mobile phone that can surf the internet anytime, anywhere, also brings convenience to life. For example, in shopping, it is not only possible to choose from physical stores, but also online. When purchasing goods, most people use online comments as a reference to judge the quality of products. Not only for shopping, but in other ways as well. Everyone is used to looking for solutions to difficult and complex problems in life through the Internet. In recent years, the epidemic has heat up, accelerating the development of the internet. Many people pass the time by watching dramas, movies, and variety shows at home. During the decision-making process, choices will be made based on comments. Common movie reviews include Yahoo! Kimo Movies, IMDB, Rotten Tomatoes and other review sites. These comments have personal emotions and subjective ideas that can be used as a reference. This research will use web crawlers to crawl the comments on the annual list of Yahoo! Kimo Movies. And use natural language processing to convert the collected data into language that machines can understand, as well as related techniques of emotion analysis to classify opinions, evaluations, and emotional trends in the comment content. The LSTM and BERT classifiers used in machine learning were compared under different influencing factors. Classify comments into three categories: positive, intermediate, and negative, and observe the performance and differences of the model in training and testing. The research results show that BERT's experimental results are better than LSTM under longer execution times, with an accuracy rate of 94.47%. Considering the factors of reliability, validity, and accuracy of movie reviews, this study suggests using BERT for data classification in movie review classification to obtain more accurate results. |
本站僅提供期刊文獻檢索。
