Activity-Trip Based Model for Friend Recommendation with Transit Smart Card Records

Abstract

How you travel, where, when, and what you do could indicate who you are. This paper discovers a possible social network between public transit passengers and develops a location–time–activity-based friend recommendation (LTAFR) model based on trips and activities of the passengers. First, trips and activities of passengers are reconstructed from the smart card data. Second, the similarity between passengers is measured in two steps for the activity similarity and trip similarity. The activity similarity is measured considering three dimensions of activity (location, time, and type). The trip similarity is measured considering both spatial and temporal dimensions. Third, a similarity score is defined as the multiplication of the activity and trip similarity values. To discover mutual relations between the passengers, the cosine similarity index is used. Finally, connected Top-k passengers are recommended as potential friends based on the highest cosine similarity values. The proposed model is implemented on a one-day smart card dataset from Brisbane, Australia. Also, the model is implemented on a household travel survey (HTS) dataset for comparing sociodemographic attributes of the recommended passengers. In the end, further investigations show that recommended potential friends have close sociodemographic attributes.