A Combination of Sentiment Analysis Systems for the Study of Online Travel Reviews: Many Heads are Better than One

Miguel Á. Álvarez-Carmona, Ramón Aranda, Rafael Guerrero-Rodríguez, Ansel Y. Rodríguez-González, A. Pastor López-Monroy


This study presents an analysis of the Rest-Mex forum task 2021, which is the first international evaluation event using tourism-related (Online Travels Reviews - OTRs) data from Mexico.  In that forum, 14 specialized sentiment analysis systems were presented. The main contribution of this research is a method to successfully combine those 14 systems specialized on sentiment analysis systems for OTRs. The outputs of those 14 systems were used to evaluate the proposed combination schemes. The systems were trained and tested with 7,413 OTRs from the city of Guanajuato, Mexico, a well-known cultural destination. All of them were collected from TripAdvisor.  We propose three schemes to combine the systems to predict the polarity of OTRs efficiently.  The combination based on deep learning improves significantly each of the results obtained in the sentiment analysis systems at the individual level. Also, the results were improved for 4 out of the 5 polarity classes in the collection. To the best of our knowledge, this is the first paper that reports results from the combination of different specialized systems in sentiment analysis for OTRs.


Sentiment analysis, OTRs, merge systems, deep learning, mexican tourism

Full Text: PDF