Spot Recommendation Through Integrated Analysis of GPS Log and Browsing Log

Abstract

This paper proposes a method to recommend spots according to user preference inferredfrom the results of integrated analyses of GPS log and browsing log. Many content-searchor recommendation services are now available to handle the large volume of contents heldby the Internet. Typically, in the absence of very precise user requests, these services needlarge amounts of browsing or search log data before they can offer suitable contents. This iscalled the cold-start problem. Of particular note, it is extremely difficult for spot recommendationservices to get adequate amounts of data. For example, users do not use restaurantrecommendation services for regular meals taken at home. Our approach is based on thefact that most portable devices, including mobile phones, have GPS and it is easy to accumulatea GPS log. The proposed method computes distributions of tags for each user usingGPS log of the user and the spot-data gathered from the Internet. The method analyzes auser’s GPS log and browsing log with the distributions. The method is based on statisticalhypothesis testing and uses Stouffer’s Z-score method for the integration. We evaluate themethod through realistic GPS and browsing logs.