Proposing sentiment analysis model based on BERT and XLNet for movie reviews

Movie reviews are a valuable source of information for potential viewers. However, reading all of the reviews can be time-consuming and overwhelming. Summarizing all of the reviews will help you make the correct choice without wasting time reading all of the reviews. Sentiment analysis, or opinion mining, can extract subjective information from movie reviews, such as the reviewer's overall opinion of the movie, its strengths and weaknesses, and the reviewer's recommendations. This information can help potential viewers make informed decisions about whether or not to watch a movie. XLNet and Bidirectional Encoder Representations from Transformers (BERT) are pre-trained advanced language models that learn bidirectional relationships between words, improving performance on many natural language processing tasks. BERT uses a masked language modeling objective, while XLNet uses a permutation language modeling objective. This experiment is based on the proposed method for XLNet and BERT, two advanced techniques and popular baseline techniques using the Internet Movie Database (IMDB) Dataset of 50K reviews and the Rotten Tomatoes dataset. We pre-processed both datasets using data cleaning, the removal of duplicate reviews, lemmatization, and handling of chat words to improve baseline technique results. The results indicate that XLNet achieved the highest accuracy on both datasets. As a result of the research experiment, sentiment analysis provides insights into how emotions and attitudes are expressed in movie reviews that can be used to predict a movie's performance based on their overall sentiment.