Social media and urban mobility: Using twitter to calculate home-work travel matrices

Highlights

• Twitter data is used to study home-work urban mobility in Madrid Metropolitan Area.

• Home and work places detection have been improved using Land Registry (Cadastre) maps.

• Origin-Destination matrices were designed and expanded using official different data sources.

• Results were validated via comparison with an official travel Mobility Survey.

Abstract

The proliferation of Big Data is beneficial to the study of mobility patterns in cities. This work investigates the use of social media as an efficient tool for urban mobility studies. In this case, the social network Twitter has been used, due to its wealth of spatial and temporal data and the possibility of accessing data free of charge. Using a database of geotagged tweets in the Madrid Metropolitan Area over a two-year period, this article describes the steps followed in the preparation and cleansing of the initial data and the visualisation of the results in Geographic Information Systems in the form of home-work matrices. The Origin-Destination matrices obtained were then compared with the official data provided by the Madrid Transport Consortium from the 2014 Synthetic Mobility Survey. The results of this comparison demonstrate that the level of precision offered by Twitter as a source of geographic information is adequate and efficient, thereby permitting a more in-depth analysis of flows between different zones of interest in the study area.

Introduction

The monitoring company StatCounter recently revealed that, for first time since 1980, Android has replaced the Windows operating system as the main internet access mode (Simpson, 2017). The virtual world is entering a new era, in which mobile phones are displacing computers as the main means of interacting with society online. Enormous growth in the use of smartphones affects the quantity of information being generated. In an increasingly technological society, with an ever-greater use of internet, nearly all large cities' inhabitants generate a digital footprint of their activities and movements, a digital trail that can be followed (Blanford, Huang, Savelyev, & MacEachren, 2015). One of the consequences of this digitalization of society is the prominent role of so-called Spatial Big Data: large quantities of spatial information that can be captured, communicated, aggregated, stored, and analysed (Manyika et al., 2011). Shadows of this digitalization are intimately intermingled with offline, material geographies of everyday life (Jin et al., 2017). This new geographic data is produced constantly, in real time, can be acquired easily at a low cost, and can be incorporated and analysed in Geographic Information Systems (Osorio & García-Palomares, 2017).

Internet users are no longer mere passive recipients of information, but have become producers of vast amounts of data, particularly through social networks (García-Palomares, Salas-Olmedo, Moya-Gómez, Condeço-Melhorado, & Gutiérrez, 2018). These social networks enable information to be created, stored, shared, and exchanged with other users, and generate a huge volume of data every day (Cao et al., 2014). Social media data have been used to study the shape of urban agglomeration based on people's activity (Zhen, Cao, Qin, & Wang, 2017), and are useful to analyse urban structure and related socioeconomic performance (Martí, Serrano-Estrada, & Nolasco-Cirugeda, 2017; Shen & Karimi, 2016; Zhang, Zhou, & Zhang, 2017). Twitter, an application based on sending short texts or other multimedia content, stands out among the most-used social networks. Geo-located Twitter is a free and easily available global data source that stores millions of digital records on human activity in space and time (Hawelka et al., 2014). According to the company's data, in 2017, approximately 500 million tweets were sent every day all over the planet. According to Twitter's own data,¹ the network has a base of 328 million active users per month, of which 82% generated information from their smartphones.

The new data sources provide information that is extremely useful in mobility studies. If we know the changing location of each person based on their digital footprint, we can analyse their general mobility patterns. Gathering geospatial data provides a unique opportunity to gain valuable insight into information flow and social networking within a society (Stefanidis, Crooks, & Radzikowski, 2013). Evolution of transport-demand model techniques have developed the need for a high-resolution database, with aggregated economic and sociodemographic attributes for daily travel behaviours modelling (Rashidi, Abbasi, Maghrebi, Hasan, & Waller, 2017). In this respect, Twitter is a particularly valuable source of data for the study of mobility, due to its ease of use on any mobile device, and because it offers the possibility of geotagging messages. Thus, the tweets of users who activate the geolocation service on their accounts, contain, in addition to semantic and temporal information, information about the geographic location from whence the tweet has been sent. It is possible to carry out studies on population distribution at any time (see, for example, Ciuccarelli, Lupi, & Simeone, 2014; Longley & Adnan, 2016, or García-Palomares et al., 2018).

Data from social networks, and in general, most of the data from sources associated with Big Data, are not data created to analyse urban mobility. This is common in the use of social networks in disciplines such as urban planning and mobility. Consequently, the use of social network data to study mobility bears great weaknesses in comparison with data from sources specifically created for this purpose, which are normally household mobility surveys. One of the conditioning factors is the importance of cleaning and pre-processing processes for data, so that the database that is finally used solely contains data from whence reliable mobility information is obtained. Another conditioning factor to bear in mind is data limitation. Thus, social networks provide a low temporal resolution, which is the average time between each tweet sent by a user. This makes it difficult, for example, to obtain data from activity sites related for different reasons to the residence and the workplace. The low temporal resolution also compels data to be collected from much greater time periods. With telephony, we can normally work with 2 or 3 months. However, with social networks, we need greater period samples (for example, 1–2 years). On the other hand, while mobile telephone data refer to very large user samples, samples are smaller in social network data, and are also more biased, since users of these social networks are concentrated into certain sociodemographic groups. Finally, using social network data for a purpose such as studying mobility requires a results validation process. Here, we have made an attempt to insist on this phase, and to validate results at different spatial aggregation scales, in order to see how far we can go with these data in obtaining travel matrices.

In studying urban mobility, the ability to generate Origin-Destination (OD) travel matrices is a fundamental tool for carrying out diagnostics, predicting demand for travel and contributing to modelling of transport and optimisation of network usage (Gao et al., 2014). OD flow matrices provide useful information for ridership forecasting, service planning, and control strategies: for modelling purposes, origins and destinations of movements and modal preferences are needed (Rashidi et al., 2017). Traditionally, household surveys or traffic counts were used to obtain travel matrices. However, these types of sources are expensive, time-consuming, static, and use small population samples (Iqbal, Choudhury, Wang, & González, 2014). Social networks in general, and Twitter in particular, provide new sources for estimating OD travel matrices. They enable us to work with high spatial-temporal resolution data, to access larger population samples, and to obtain this data free of charge (when downloaded via streaming). The aim of this paper is to validate the use of Twitter data in assessing urban mobility patterns by using the Madrid Metropolitan Area as case study. Trips from home to workplaces were chosen as a study motive. The economic value of the workplace is not of academic interest alone, but also of practical interest, as they are key spaces for urban design and planning. Understanding how workers move throughout the day is important to design and manage transport systems and public spaces (Pajević & Shearmur, 2017).

Previous studies focused on obtaining OD travel matrices through new data sources mainly used phone data, so-called call detail records (CDRs) (Chen, Ma, Susilo, Liu, & Wang, 2016). However, mobile phone companies are extremely reluctant to hand over their data, and if they do, charges are substantial. Furthermore, information from CDRs is usually linked to the coverage range of the antennae, which means that one frequently works with Voronoi. Using CDRs, spatial resolution is occasionally lower, and they frequently do not match the distribution of land uses or transport zones which are necessary for an accurate modelling of future situations (Järv, Tenkanen, & Toivonen, 2017). Bearing these two main limitations of previous studies based on phone data in mind, the main contribution of this paper is to validate Twitter as an alternative source to phone data in estimating OD travel matrices and detecting home-work trips, benefiting from: a) free-of-charge data; b) the availability of geolocated tweets and the possibility of spatially joining them to transport zones (which are of great interest to transport managers).

When processing Twitter data to obtain the matrices, some methodological improvements were carried out, as opposed to previous works. Thus, information on land uses from Land Registry (Cadastre) maps was used. These maps benefit from high spatial detail, offering greater precision when it comes to detecting home and work places. Furthermore, various factors of expansion were evaluated for the sample. On one hand, the authors worked with expansions based on the origins of the trip, using population data according to place of residence from official sources (Census). On the other hand, the matrices were expanded using data relating to the destination of the trip, using information about the workplace (from National Institute of Social Security records). The estimations were carried out at two different scales of analysis: a) municipality level (with finer spatial detail); b) metropolitan-zone level. The results were validated via comparison with the data from the Synthetic Mobility Survey carried out by the Madrid Transport Consortium in 2014, thereby enabling us to evaluate the quality of the matrices obtained by Twitter data and the sensitivity thereof, both to the type of expansion factor used and to the degree of spatial desegregation.

This article is divided into six sections. After this introduction, Section 2 reviews the research carried out to date using Twitter and OD matrices. Section 3 defines the study area, and Section 4 introduces the data and sources we used. Meanwhile, Section 5 reviews the methodology used in the research. The obtained results are analysed in Section 6 to establish a series of conclusions in Section 7.