Using data mining on linked open data for analyzing E-procurement information
Eneldo Loza Mencía, Simon Holthausen, Axel Schulz, Frederik W. Janssen · 2013
Abstract Understanding complex procurement information landscapes and exploring how procurement information can be used to support strategic decision-making is important with the increasing amount of information available in the WWW. In this paper, we cope with this challenge and describe how data mining techniques can be applied on semantically linked data to estimate the number of bidders in public contracts. We introduce a general approach in order to convert linked data in a relational format which can be used by traditional machine learning approaches. Afterwards, we apply common techniques such as discretization, processing of text fields, feature selection, state-of-the-art machine learning algorithms, and more. Alternatives were evaluated and compared. We estimate an accuracy of 32.69 % (with a baseline of 30.34%) and costs of 0.37 (baseline: 0.49) for our best configuration, a cost sensitive ensemble of classifiers, for the DMoLD’13 competition. 1