Sentiment analysis on Vietnamese text: A fine-tuning based approach and an evaluation tool

Abstract

Sentiment analysis on informal text on the Internet has received considerable attention be- cause of the growing large amount of text online, and the value it brings to business and social organizations. This thesis aims to review all the approaches to sentiment analysis on Vietnamese text on social media and study the potential of transfer learning as a solution to the task by (1) introducing a procedure to apply transfer learning to sentiment analysis on Vietnamese informal text, (2) conducting experiments accordingly, and (3) reasoning from the results. Besides, another purpose of this study is to propose a tool to quickly evaluate algorithms for sentiment classification on Vietnamese text in general. Experimental results in this study suggests that transfer learning is promising in solving sentiment analysis tasks on Vietnamese informal text and deserves further exploration in future works. Keywords: sentiment analysis; transfer learning; BERT.