Aspecta: Improving Public Procurement using Natural Language Processing

Public procurement is a vital process that governments undertake to obtain goods and services needed to deliver public services efficiently. However, the procurement process is often tedious and time-consuming, with many challenges faced by both government agencies and suppliers. 

The procurement process utilizes software support at many steps. Data-wise adoption of graphical user interfaces of applications and tools can potentially reduce time on task and error rate, if designed correctly. Recommendation, suggestion or validation components can significantly help users to provide consistent and error-prone information. Data provided will be of higher quality.KInIT teamed up with Aspecta to help them pursue their goal of supporting digital transformation. We will research efficient methods for automated text classification in procurement requests. We will carry out a set of experiments that will result in a report on suitability of multiple approaches to multilabel classification. We will explore linguistic and statistical ones, including incorporation of large language models, such as SlovakBERT, that are both popular and well performing on many NLP tasks. Based on the results, adaption of the existing software application for procurement requests may significantly improve in efficiency.