Abstract
Sentiment classification is a crucial problem in natural language processing and is essential to understand user opinions. There are two main approaches to solve this problem, one is the classified-based method, the other is the lexicon-based method; however, both methods perform not well on the long-sequence methods, and each method has its advantages and disadvantages. This paper introduced a new method called Lexiconed BERT, which cream off the best and filter out the impurities from the above two methods. The evaluation shows that our model achieves excellent results in the long sequence sentence and reduce resource consumption significantly.
Highlights
Sentiment Analysis, drawing the sentiment polarize from the given sentence, is an essential and practical application in Natural Language Processing
There are mainly two general methods used for sentiment analysis: the classified-based method which regards the sentiment analysis as a classification problem, and trains a classifier (SVM, RNN, etc.) to make the prediction; the lexicon-based method which used the sentiment lexicons---the words labeled with specific sentiment tag by hand, predicts the sentiment based on an algorithm
The sentiment benefits from both methods: to classified-based method it could be flexible due to the various selected features, but at other times it could be more accurate on some certain corpora domains by lexicon-based methods
Summary
Sentiment Analysis, drawing the sentiment polarize from the given sentence, is an essential and practical application in Natural Language Processing. The sentiment benefits from both methods: to classified-based method it could be flexible due to the various selected features, but at other times it could be more accurate on some certain corpora domains by lexicon-based methods. Our contributions include: We put forward a new pre-trained task for BERT, that a task cuts out for the sentiment analysis task, which does not predict the masked words but the polarization of the sentiment lexicons; We take advantage of the lexicon-based method, which helps us to select useful features and accelerate both the training and inference speeds. Compared with Lexicon-based methods, our model only needs basic sentiment lexicons to be tagged and updated
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
