Temporal Signature for Location Similarity

Abstract

An increasing amount of user data, e.g., check-in history, from location-based social networks has become available for recommending new places. Recently, temporal check-in information was taken into consideration and has shown promise to improve the performance of current location recommenders. In this work, we study whether the visit time of a location can reflect the nature of the place and can be used to measure similarity between locations. In particular, we consider a new location feature, temporal signature (TS), to capture the temporal visit patterns of the location by aggregating all users' data, and apply various time series distance measures. We design several empirical studies with real-world data to evaluate the goodness of TS. The results show that TS features reflect the location semantics, geospatial locality, and location/category similarity in time.