Text Preprocessing Audit Findings of Financial Statements: Preparing the Data for Further Analysis

Audit findings stem from the audit process conducted by the DKI Jakarta Province Representative of the Supreme Audit Agency. To provide relevant information for decision-making, audit findings are grouped based on the problems encountered during the audit. Audit findings are often presented in unstructured and diverse formats, making them challenging to use directly for text classification or further analysis. The raw text must first undergo comprehensive preprocessing to ensure these findings can be effectively processed. This study focuses on the various preprocessing techniques required to transform complex audit findings into standardized, clean, and structured data that can be used in future stages of analysis, such as fine-tuning models. The preprocessing steps include cleaning text, removing irrelevant characters, symbols, and stop words, case folding, and tokenization to split sentences into manageable units. This approach aims to resolve challenges associated with inconsistent language use, domain-specific terms, and irregular sentence structures found in audit reports. The result is a structured dataset ready for application in advanced deep-learning tasks. By focusing on preprocessing, this paper highlights the critical role of data transformation in preparing unstructured audit texts for accurate and efficient classification, ensuring that the information is correctly formatted for further fine-tuning and model training.