A recommender system based on car pairwise comparisons on a mobile application using association rules

Abstract

Numerous product information mobile applications (APPs) have been developed and their download counts are not negligible. The recommendation functions of Apps will help users to efficiently find related products and subsequently the user satisfaction will increase. This study aims to analyze pairwise comparison data using association rules to help the APP developer establish the recommendation system. The data comes from the members' comparison records from a new cars database App developed in Taiwan, i.e. NewCarsDB (www.newcarsdb.com). We collected a sample of 40 car brands and 870 vehicles comparison records during 2015/1/30 to 2015/4/2 with 30,867 car pairwise comparison records. This study develops two metrics, i.e. (1) width (quantity of cars which have associated products) and (2) average depth (each car with quantity of associate) to evaluate the results of different thresholds. Results show that (1) Support adjustment has influence on width; (2) The confidence adjustment under thresholds lower than 10% has little impact on width but their impact on the average depth are not negligible. The results can be used as references for associating products and can also be used in recommending a new product to potentially interested members. Moreover, the members of new cars database App will have better experiences whereas the potential market to improve advertising effectiveness can be developed at the same time.