A Study on Long Short Term Memory Algorithm in Sentiment Analysis with VietNamese Language

Abstract

The increase in the number and content of electronic information sites such as Facebook, Twitter, Instagram,… is not only a place to provide information in the form of events but also a place where users express their feelings and exchange information, feelings, experiences about life issues. Research in content mining comments to observe user reactions to make adjustments and improvements in organizations is being studied. However, how to build a good prediction algorithm on the Vietnamese data set is not simple because of Vietnamese polysemy and polymorphism. In this paper, the authors have developed the Long Short Term Memory algorithm - an extension of RNN to learn sequential data and long-term connections more accurately than RNNs standard. Using a dataset of Vietnamese comments from social networks collected by the research team together with a preprocessing, standardization, and labeling solution. The experimental results indicate the proposed model achieves an accuracy of 98.52%. From this experimental result, it is possible to further develop the research on the emotional analysis of Vietnamese sentences and can be applied in practice.