A Recommendation System for Shared-Use Mobility Service through Data Extracted from Online Social Networks

• Main Authors: Renata Lopes Rosa, Eduardo Lucio Lasmar Junior, Demóstenes Zegarra Rodríguez
• Format: Article
• Language: English
• Published: Croatian Communications and Information Society (CCIS) 2018-12-01
• Series: Journal of Communications Software and Systems
• Subjects: Recommendation system; shared mobility; mobile applications; online social networks; machine learning; social web analysis tool
• Online Access: https://jcomss.fesb.unist.hr/index.php/jcomss/article/view/602

In recent years, the shared mobility service hasincreased in many countries across the world because its low cost and several shared-use mobility applications on mobile devices. Commonly, if a ride is shared between people with similar preferences, users likely feel both more comfortable and safer.In this context, the main goal of this article is to classify userswith similar preferences, in automatic manner, to improve user’s quality of experience in ridesharing service. To obtain initial data, subjective tests are carried out using questionnaires and their results are used to determine ridesharing profiles. Then, some basic user profile information is extracted from Online Social Networks (OSN) to determine an user profile based on preferences in ridesharing service. The user profile classification is performed through different machine learning algorithms, which use as input the data extracted from OSN. Two case studies of shared-mobility are treated, (i) sharing a ride with a passenger with a similar hobby [2], and (ii) sharing a ride with people thatsupport an opposite football teams. In this work, a novel contribution is the use of Hybrid Discriminative Restricted Boltzmann Machines (HDRBM) technique for classification, which results overcomes other algorithms, such as Random Forest, SVM and DRBM. The experimental results presented a correctly classified instance of 96:9% and 97:3% for the cases of sharing a ride with people with similar hobby and support different football team, respectively. Finally, a Recommendation System (RS) is proposed, which efficiency is compared with a basic RS, obtaining a Pearson correlation coefficient of 0:97 and 0:79, respectively.