Avtomatizacija delovnih tokov nabave v gradbeništvu

Alič, Maks (author), Klinc, Robert (mentor), Brelih, Anja (comentor)

Magistrsko delo obravnava optimizacijo procesa nabave v gradbeništvu z avtomatizacijo obdelave popisov in iskanja podizvajalcev. Zaradi neenotne strukture podatkov in pomanjkanja standardizacije v praksi raziskava temelji na analizi realnih gradbenih popisov ter reševanju izzivov pri obdelavi vhodnih podatkov. Za sistematično vrednotenje smo razvili lasten testni popis. V delu smo preizkusili in primerjali več metod za klasifikacijo postavk gradbenih del v kategorije: regularne izraze s ključnimi besedami, lematizacijo, vektorizacijo besedila in kosinusno podobnost ter velike jezikovne modele. Ob tem je bil razvit tudi sistem za pridobivanje podatkov in oblikovanje nabora podizvajalcev glede na vrsto del in geografsko lokacijo. Metode klasifikacije in iskalnik podizvajalcev smo najprej preverili na razvitem testnem popisu ter po vseh 12 statističnih regijah v Sloveniji. Na podlagi teh ugotovitev smo izdelali prototip spletne aplikacije. Delovanje celotnega sistema je bilo preizkušeno in ovrednoteno na treh realnih študijah primerov (Dom starejših občanov Dragomer, Vrtec Mežica in Hotel Križnič). Rezultati potrjujejo, da iskalnik podizvajalcev deluje dobro in učinkovito optimizira proces pridobivanja ponudb. Nasprotno pa so rezultati klasifikacije postavk slabši, kar jasno izpostavlja nujno potrebo po standardizaciji tako oblike kot tudi vsebine popisov gradbenih del v Sloveniji.

• Language: Slovenian
• Keywords: digitalizacija gradbeništva, popis del, klasifikacija besedila, iskanje podizvajalcev, standardizacija podatkov
• Work type: Master's thesis/paper
• Organization: FGG - Faculty of Civil and Geodetic Engineering
• Year: 2026
• PID: 20.500.12556/RUL-186274
• Publication date in RUL: 29.08.2026

Secondary language

• Language: English
• Title: Automation of procurement workflows in construction

Abstract

This master's thesis addresses the optimization of the procurement process in civil engineering by automating the processing of bills of quantities and the selection of subcontractors. Due to non-uniform data structures and a lack of standardization in practice, the research is based on the analysis of real construction bills of quantities and on addressing challenges in processing input data. For systematic evaluation, we developed our own benchmark bill of quantities. In this work, we tested and compared several methods for classifying construction work items into categories: regular expressions with keywords, lemmatization, text vectorization with cosine similarity, and large language models. Additionally, a system was developed to retrieve data and form a pool of subcontractors based on the type of work and geographic location. The classification methods and the subcontractor search engine were first verified on the developed benchmark bill of quantities and across all 12 statistical regions in Slovenia. Based on these findings, we developed a prototype web application. The performance of the entire system was tested and evaluated on three real-world case studies (Dragomer Home for the Elderly, Mežica Kindergarten, and Hotel Križnič). The results confirm that the subcontractor search engine performs well and effectively optimizes the tendering process. Conversely, the classification results for work items were less effective, highlighting an urgent need for standardization in both the format and content of construction bills of quantities in Slovenia.

• Keywords: construction digitalization, Bill of Quantities (BoQ), text classification, subcontractor sourcing, data standardization