Applications of Artificial Intelligence for Auditing and Classification of Incongruent Descriptions in Public Procurement

Abstract

Context: Despite the advancement of technology, many services and information systems, especially in the public sector, still use unstructured natural language descriptions of products, services, or events, making their classification and analysis difficult. For efficient audits, it is necessary to classify and automatically totalize invoices issued for the purchase of products, considering their unique identification codes. Problem: The codes are not always registered correctly by the suppliers. Furthermore, if the product description is considered an alternative to the code, as aforementioned, this is not a uniform field, having free and variable writing. Solution: This work aimed to identify and characterize the approaches, techniques and intelligent algorithms used to classify incongruous textual descriptions present in the invoices issued. IS theory: General systems theory; Competitive strategy (Porter); Knowledge-based theory of the firm. Method: A systematic mapping was conducted to find the primary studies in the literature and collect evidence for directing future research. Summary of Results: 225 articles were identified, with Scopus and Web of Science being the bases with the most articles. Only 15 articles passed the inclusion and exclusion criteria. Among the approaches used, supervised machine learning stands out, present in 60% of the works. The most widely used techniques were Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), present in 40% of the articles. Contributions and Impacts in the IS area: The research showed that the use of artificial intelligence techniques helped to mitigate the problem of classification and analysis of invoices with incongruous codes and descriptions, which can help in the audit process, investigation, and fight against corruption. Finally, trends and gaps to be explored were also presented.