Predicting the number of bidders in public procurement

Public procurement constitutes an important part of economical activities. In order to effectively use public resources, increasing competition among firms participating in public procurement is essential. In this work, we investigate the impact of content information on the number of bidders in public procurement. We explore 6 different groups of features including n-grams, named entities, language of notices, country of the authority, description length, and CPV codes. In our experiments, we show that our proposed models outperform all baselines. In particular, k-nearest neighbor model with n-grams achieves the best prediction accuracy. Our model can be used by public procurement officials to automatically examine procurement notices and detect the ones causing low competition. Besides, participating firms can use our model to predict potential competition they will face, and make better decisions accordingly.