감사보고서의 비정형 내용분석과 감사보수 및 시간을 이용한 감사의견의 적정성 연구 : 텍스트 마이닝과 감성분석 기법 적용을 중심으로

Currently, a majority of domestic audit reports convey qualified opinions, which raises serious questions about whether audit reports provide appropriate level of decision support power to interested entities. This study starts with this research issue. This study adopts big data-driven text mining and sentiment analysis methods. In particular, all the relevant words used in the audit reports are extracted by applying text mining method. Next, based on BoW (Bag-of-Words), sentiment analysis is applied such that each word induced from the BoW method is given its proper emotional score ranging from +3 to -3 in line with 5 experts' opinions. Based on the emotional scores of those words extracted from audit reports, the sentiment index of audit report is computed. Our main hypothesis is that if audit reports' sentiment scores describe audit risks using words with negative emotional score, audit fees and hours would increase as the sentiment index of audit report decreases. By using KOSPI listed firms with qualified audit opinion in year 2015-2017, we document that audit fees and hours are negatively related with the sentiment index of audit report. These findings show that auditors charge higher audit fees and put more hours on risky clients and reveal these risks through audit reports by choosing words with negative emotional score.