Yesterday I had a wonderful conversation with Jacques Lévy and André Ourednik (of the Choros laboratory, EPFL) on the methodology to explore a dataset of “spatialised messages”. We drafted a couple of points and ideas that I am trying to summarise in this post.

1. Geometry and Geography are not synonyms. This is the main point that Jacques raised as often these two dimensions are conceived as the same thing. Geometry might be a good approximation of geography under exceptional circumstances. For instance, if a part of the city has a linear structure and the density of buildings is constant and the transport systems are available in a uniform way on the territory, and etc. So in this case we can use this approximation. However current approach to spatial cognition tend to move from the assumption that geography is Euclidean. If people have distorted perception of this “reality” then they are wrong and these false perception should be treated as “mistakes”. Of course Jacques and I agreed that we can consider different starting assumptions lie the fact that the geography of a place is given by these distortions which are not mistakes.

2. On the pragmatical approach on the analysis of the messages, Jacques proposed to measure the messages’ geometrical average distance against the threads’ average connection distance. This measure might give us a view on the proportion of conversations that extend beyond the local clusters to represent similar features of the city on different neighborhoods.

3. A second idea emerged in the meeting is that of comparing the messages clusters (or the messages distribution) with other census information on the same area of the city. For instance: does the highest density of the points correspond with the highest density of shops in the area? Does it matches with the area that the users uses the most for various reasons (i.e., work, leisure, residence)? An operational exploration that we agreed to perform on the data is to confront the messages distribution with different layers that we can obtain from statistical bureau or census data.

4. Finally André Proposed to work on a possible anamorphosis map considering different attributes of the messages, as for instance the semantic distance between each pair of points. Other dimensions can be explored as the time evolution of the conversations, the threading between the messages or the social network that developed between the users of the system. The basic principle of the anamorphosis is that the map is divided in cells. For each morph a feature is chosen and measured in each cell. Then the area of the cell is set proportional on the feature assigned so that the ultimate map will equalise the chosen feature across all the map.

During our last meeting with Pierre, we sketched briefly a couple of other possible explorations to perform on the dataset. Pierre was proposing to build different maps giving more importance in turn to the geographical, the semantic or the social dimension of the data. Then for each map we could measure a couple of parameters to confront like the dispersion (are the messages more dense or sparse); the structure of the connections (what is the form of  the links of the Minimal Walking Tree built on the map); number of links per node (again with the MWT); etc.

We ended the meeting with the proposition of exploring some of the directions highlighted.

Copyright notice: the present content was taken from the following URL, the copyrights are reserved by the respective author/s.

Tags: anamorphosis maps, map algorithms, maps, psychogeography, spatial clustering, urban exploration