Construction of a dictionary for automated discrimination of auditor’s opinion type

Abstract

This paper presents a simplified procedure for construction of a tone dictionary used for the automated recognition of the auditor's opinion type. The study was based on audit opinions prepared in the Polish language. Our sample consists of 362 reports for the period 2013-2016. It served as a basis for creating the language corpus. We applied text mining techniques combined with correlation analysis to assess information content of words on the basis of correlation with the type of auditor’s opinion. We have demonstrated that a modified auditor’s report (negative or qualified) has a significant correlation with tone while in the case of an unqualified opinion there is no such correlation. The procedure of dictionary creation and the dictionary itself can be used to develop automatic trading algorithms.

Keywords

• audit opinion
• sentiment
• tone
• algotrading
• machine learning