Showcasing research in data mining is a challenging topic. Developed models of everyday phenomena are often only understandable by domain experts which might lead to low general adoption. We demonstrate an interactive dashboard to visualize the complex domain of international travel over time. Given that spatio-temporal phenomena such as mobility can not efectively visualized using consumer software, we developed a web-based system, where users can explore how the behavior of international and domestic travelers changes over time and easily create their own analyses.
Human mobility analysis is a widely researched topic with many
facets such as next location prediction [
The underlying data stems from a self-collected data set from
Twitter. Due to that most content of Twitter is available for the
general public, it has been frequently used as a source for
researching human mobility [
Dashboards provide users with a customizable overview about the state of a system without requiring them to work with the raw data and using programming techniques. Thus, they are an ideal tool for the general public to explore complex topics such as stock markets, misinformation on social media1 or global pandemics2.
In this position paper, we present our initial prototype for a
traveler mobility dashboard based on mobility data derived from
social media. We present the motivations and design considerations
to build a dashboard that can visualize analyses concerning trends
1http://csmr.umich.edu/projects/ify-quotient/ [
Mobility data about individuals is valuable, since it can be used for
many purposes and it is quite complicated to collect it on a larger
scale. Most data sets are limited to national boundaries, since
traditional methods of collection, such as mobile phone communication
records [
Whenever a user tweets with the geolocation option enabled, her approximate location is attached to the message. The downside of LBSN mobility data is that it gives an incomplete picture of the user’s mobility — the location is only known when she decides to tweet. This makes it necessary to assess the quality of the data with respect to the use case, eventually, discarding trips of insuficient data quality.
Feature Tweets with geolocation Users International Travelers Domestic Travelers Trips International Trips Domestic Trips Countries visited Observation period
Value
Since 2018, we collected timelines of Twitter users and using the Python tripmining library3, we segment geolocated tweets of a user’s timeline into periods of being at home and on travel. If a user tweets at a relatively constant pace, this data can be quite reliable, which is also one of the of the library’s features to assess using various metrics.
Table 1 summarizes the statistics of our Twitter trips data set used for the dashboard. As the data collection is ongoing, these 3https://github.com/LinusDietz/tripmining Lukas Vorwerk and Linus W. Dietz specific chart areas. To the best of our knowledge, this platform is the first ofering an interactive analysis of global travel trends. numbers are constantly growing. While only parts of the data set are currently publicly available4, we are able to provide mobility data to interested researchers on a trip-level. 3
The dashboard consists of three interactive visualizations that invite users to engage with the data and create mobility analyses on their own. The central feature is a choropleth world map that reflects the number of inbound or outbound trips on a country-level basis. The exact number is shown when hovering over the respective area on the map. By using a toggle switch, the user can decide whether to display inbound or outbound trips. Upon clicking on the area of a specific country, users are presented with a detailed analysis of this country. They can now determine where the residents of this country prefer to travel to and from which other countries travelers visit this country in which quantities. Figure 1 shows the number of outbound trips in our database undertaken since 2015.
Besides the map, a list containing the top ten countries for the selected scenario is displayed. An additional visualization in the form of a line chart displays the mean and median trip duration per year or month. To provide insights into the data the visualizations are based on, the dashboard also includes the statistics shown in Table 1 and a chart that indicates the number of trips per year or month (see Figure 2). Furthermore, a time slider ofers the possibility to uncover temporal trends over the last decade. All of the described dashboard elements are interconnected, i.e., if the user selects a specific country or year, all visualizations are updated accordingly.
The dashboard is developed using Dash5, a “Python framework
for building web analytic applications”. The usage of Dash enables
further interactions with the visualizations such as zooming in on
4https://github.com/LinusDietz/JITT2020-Mining-Trips-Replication [
Given the intrinsic properties of the data, some analyses work better than others using the dashboard. What works quite well is to get an indication of the popularity of a destination country as well as the popular travel seasons aggregated on a monthly basis. Certainly, the raw numbers have little meaning, since they only account for the mobility of Twitter users. The relative popularity of countries in the Western World, however, should be meaningful. On the contrary, the numbers for countries where Twitter is not much adopted, e.g., due to state censorship, are certainly not representative. Furthermore, the overall amount of data might be enormous (cf. Table 1), but when subdividing it into 7 years (currently about 73 months), 241 countries, and inbound and outbound trips, the amount of data points can become quite low, especially for small countries. Finally, it should be noted that the data collection is slower than its creation, and it takes time until enough recent trips are collected. For this reason, the most recent data is the most incomplete, since trips that are currently occurring can not yet be included in the data set.
Therefore, the target audience of the dashboard is rather the general public than researchers. The number of possible analyses is intentionally limited at the moment, since we believe that increasing complexity would hinder layman to interact with the dashboard to produce their own simple, but useful analyses. 5
This paper described the interactive visualization of international traveler mobility from Twitter. The underlying data has been collected over the last year and shows the high value of LBSN data for such analyses. With the presented dashboard, we provide a tool that can be used to perform own analyses on the complex domain of international travel. Currently, the dashboard allows to explore the temporal variability of the number of inbound and outbound trips of Twitter users.
In the future, we plan to improve the granularity of the analysis
into smaller regions, such as federal states or touristic regions [