Preference aware route recommendation using one billion geotagged tweets

Abstract

Twitter is a popular social networking service where people send short messages called tweets. Tweets contain metadata such as language, hashtags, geotags, and time of creation. We focus on the geotags of tweets. A Geo-tag is georeferenced information that indicates the geographical origin of a tweet. Geotagged tweets provide an excellent opportunity to understand the underlying user behavior. We propose a preference-aware route recommendation method relying on over one billion geotagged tweets. The method can recommend routes based on user preference by extracting a subset of one billion geotagged tweets according to user preference and using that subset to generate a cost function for route discovery. The proposed method assumes that areas with a high density of geotagged tweets are areas of high interest. In other words, if the density of geotagged tweets with user preference is superimposed on the cost of the route search, the users' preference can be considered when recommending a route. We highlight a nighttime route recommendation mechanism for a case study of our method. We hypothesize that geotagged tweets sent out at night indicate human activity at night. In other words, areas with a high density of geo-tagged tweets are considered to be areas that are vibrant at night. In addition, it is empirically clear that nighttime vibrant is also based on brightness. Therefore, we utilize nighttime tweets and nighttime light data to recommend routes. We extract a subset by calculating nighttime from tweet metadata. Tweets data are divided into grids and used to calculate a vibrant grid from a weighted tweets grid and a nighttime lights grid. Edge is weighted from vibrant cell values and road network edge lengths to recommend a vibrant route based on weighted road network edges. We experimented in Shinjuku, Tokyo, Japan, between two stations. As a result, based on the objective evaluation, we recommended a vibrant route.