Developing an Emotion Estimation Model Using Tweet Data

Abstract

This study creates a model for estimating sentiment from tweets that include “work style reforms “.

The system to extract sentiment scores from the tweets was developed using MS Excel and VBA (Visual Basic for Applications). A total of 11,272 tweets were collected using Twitter's API v2 (Application Programming Interface v2), and 8,570 tweets were selected for analysis after removing retweets. The data was divided into training and test sets, and sentiment analysis was performed. The sentiment scores, labels, and tweet text obtained from the analysis were trained using a support vector machine (SVM). The performance of the test data was evaluated using a stratified 5-part cross-validation method.

The results are as follows: The sentiment analysis results for the training and test data indicated that 2,896 and 2,824 tweets were negative, respectively, while 1,389 and 1,461 were positive. The performance evaluation of the sentiment estimation model showed an area under the curve (AUC) of 0.936.

This study developed a system using Excel to extract sentiment scores and analyzed the sentimental words used in conjunction with “work style reforms”. The study developed an sentiment estimation model with a high classification performance of AUC 0.936 based on test data validation.