2022 Volume 34 Issue 4 Pages 673-682
In this research, we propose an automatic construction method for financial sentiment lexicons on bond market. Research on financial sentiment lexicons that specialize on stock market has already been conducted, but it was not suitable for bond market. Therefore, in this study, we have developed a method to automatically build a financial sentiment lexicon that includes expressions on bond market. In addition, we succeeded in acquiring new expressions by using a tagging model for the learning data of a part of the automatically constructed financial sentiment lexicons. The evaluation of our method showed that the accuracy of expressions acquired by our method was 91.0% and the accuracy of the polarity assigned to the expressions by our method was 75%. Moreover, even when the automatically constructed financial sentiment lexicons were used as training data, the accuracy of newly acquired expressions was 80.0% and the accuracy of polarity assignment was 74%. It was shown that it is possible to acquire new expressions without much loss of accuracy.
