Estimating Travel Demand from Twitter using an Individual Mobility Model: In Sweden, The Netherlands and São Paulo

Abstract

The cost of conducting household travel surveys is increasing, while the response rate is decreasing, pushing researchers to explore new sources of data that can be used to estimate travel demand. Among these new data sources is geotagged tweets from Twitter due to its large quantity of available data and low cost of access. At the same time, using Twitter for travel demand estimation has garnered criticism regarding the biases inherent in Twitter data. This thesis uses geotagged tweets from three regions: Sweden, the Netherlands and São Paulo, to quantify the bias in Twitter data and develop a novel model that estimates travel demand by de-biasing the raw Twitter data. The model integrates two natural dimensions of individual mobility: regularly returning to habitual locations and occasionally exploring new locations. The proposed model addresses the under-representation of habitual places such as home and workplace and corrects the geotagging behavioural bias of overly representing long-distance travel. The model is validated against external data sources in each of the three regions and it is found to result in significant improvements over contemporary methods for using Twitter data for travel demand estimation. The model’s parameters are robust across regions studied, and by using the parameters found in this thesis one can expect the same improvements compared to contemporary approaches when applied to other regions.