Quality assessment of collaborative content with minimal information

Content generated by users is one of the most interesting phenomena of published media. However, the possibility of unrestricted edition is a source of doubts about its quality. This issue has motivated many studies on how to automatically assess content quality in collaborative web sites. Generally, these studies use machine learning techniques to combine large number of quality indicators into a single value representing the overall quality of the document. This need for a high number of indicators, however, has detrimental implications both on the efficiency and on the effectiveness of the quality assessment algorithms. In this work, we exploit and extend a feature selection method based on the SPEA2 multi-objective genetic algorithm. Results show that we can reduce the feature set to a fraction of 15% through 25% of the original, while obtaining error rates comparable to the state of the art.