A0341
Title: Machine learning-based sentiment analysis with fuzzy data to predict online customer satisfaction
Authors: Rosa Arboretti - University of Padova (Italy)
Elena Barzizza - Università degli studi di Padova (Italy)
Nicolo Biasetton - Università degli Studi di Padova (Italy) [presenting]
Riccardo Ceccato - University of Padova (Italy)
Marta Disegna - University of Padova (Italy)
Luca Pegoraro - University of Padova (Italy)
Luigi Salmaso - University of Padova (Italy)
Abstract: Big Data and Web 2.0 allow gathering a huge amount of free and timely online reviews that customers write on a variety of products/services. Generally, review web platforms ask customers to leave a textual review along with rates regarding the overall product/service and its key aspects. Most of the studies adopting ML-based SA use the general rating as an independent variable, and some of them also include aspect rating within the application. This type of approach makes it possible to predict the general rate through the specific rates collected on product/service aspects, if available, and textual reviews. However, despite being a user-friendly, easy-to-develop and to-administer instrument, Likert-type scales used to collect rating data are unprecise tools which generate ordinal variables that cannot be analysed by statistical methods defined on a metric space. In fact, the distance between two consecutive items cannot be either defined or presumed equal. In such a context, fuzzy theory can be used to recode customers' rates into fuzzy numbers before the adoption of a suitable ML algorithm for fuzzy data. This procedure allows the obtention of more precise predictions of general customer satisfaction. Our approach is presented and discussed using real data, highlighting its main advantages.
