In this short paper we are going to present what we have done up to now. In order to clarify the context for the reader, we give some useful links:
Here's a detailed description of the current graphs we are working on:
-
The airline routes graph. Each node represents an airport in the world: it includes the city name, the geographic coordinates of the airport and the country. Two nodes are linked if there exists a route between the airports. Each edge is weighted proportionally to the air traffic between the incident nodes.
-
The twinning-relationships graph. Each node represents a city in the world: it includes the city name, the geographic coordinates, the population and the country. Two nodes are linked if they are twinned.
-
A reduced version for each graph 1) and 2): only nodes appearing on both graphs are kept, the others are discarded (i.e. each twinned city must have an airport and viceversa)
-
Countries graphs. In addition to the previously described graphs, we decided to integrate an alternative graph structure in our analysis. For both the twinning-relationships graph and the airline routes graph we built a corresponding one as follows:
- the vertices are representative of the countries around the world;
- every edge between two countries is weighted with the number of occurrencies of edges between the two locations in the original graph.
The reason behind this choice is the will to better capture eventual relations between twinning relationships and air traffic, adopting a more macroscopic perspective. Even if this option causes us to lose information about links within a single country, it gives us the advantage of studying the case on the international level, allowing to overlook all those cases in which the air routes graph fails to represent people travels between two locations (e.g. when a traveler leaves from Amsterdam and lands at the Venice airport in order to reach Verona as the final destination by other means).
The main objective of this project has not changed: we aim to find similarities between the "sister cities" graph and the "airline routes" graph.
During the start of the project we encountered some small issues with the datasets: the first dataset is the result of a query to Wikidata Query Service, and consists of cities all over the world along with their sister cities (we filtered this set by selecting cities with population > 10000); each city is an entity identified by a code in Wikidata, which can be used to retrieve information about the city, such as its label (i.e. its real name in English, for instance London), its country, its population, etc. The other dataset comes from Openflights, and represents some common airline routes among cities in the world, which are identified by their names (while airports by their airpoirt codes). Since this dataset does not use the same scheme as the previous one, there were a series of issues, which in part have not been solved yet: we cannot compare cities using Wikidata id, so we relied just on names, which however sometimes are not always equal, due to different encodings, like Tel Aviv and Tel-Aviv or San José and San Jose. When we retained common cities, in order to compare the two graphs, we took into account these different encodings, by lower casing the strings and stripping the weird accents. Another problem of these two datasets is that common nodes are very few (~ 800) with respect to the total number of cities (~ 4500) or airports (~ 3000) (due to the fact that cities with airports do not appear very often in the first dataset).
Some nice pictures of the two datasets on a world map can be found on the repository (link above).
Perhaps (as an alternative or additional approach to the 4th point above) in future simulations we might consider a new graph, which takes into account flow of people across sister cities which do not have an airport but are close to airports for which there exists a route between them in the airline routes database: we think it could better express the relationship between air traffic and twinning partnerships.