Abstract

GONZALEZ-EVORA, Felipe  and  CENTENO-MORA, Óscar. Text minig in the National Transparency Survey 2019. Rev. Mat [online]. 2022, vol.29, n.2, pp.261-287. ISSN 1409-2433.  http://dx.doi.org/10.15517/rmta.v29i2.46379.

Coding and analyzing open-ended questions from opinion survey is often time consuming. Text mining offers an alternative for this type of problem. Data comes from the 2019 National Survey of Perception on Transparency open-ended questions. Text mining is applied from a descriptive and predictive approach: the latter has a predominant interest in performing the automatic coding of responses or categories using supervised machine learning. Support vector machine algorithms, naive Bayes classifier, random forests, XGBoost, and closest neighbors are used. The results of the descriptive analysis improve the descriptions, visualizations and relationships in the analysis of the open-ended questions. The predictive analysis reports that the algorithms with the highest selection occurrence for the open-ended questions were the naive Bayes classifier and the random forests, showing accuracies between 48% and 76%. Similar results were obtained compared with the pre-established categories. Satisfactory results are seen in the comprehensive analysis of the 12 survey

questions.

Keywords : opinion surveys; open questions; text mining; supervised machine learning..

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )