Home > Uncategorized > MIT-Reality Mining Dataset – location analyis

MIT-Reality Mining Dataset – location analyis

November 12th, 2010 Leave a comment Go to comments

I have spent some time looking at the reality mining dataset, to see how useful it is for inferring location about individuals. Firstly I generated a matrix, People over visited cell towers, which was huge, due to the large number of cell towers in the dataset (32, 000). Then I used this data to generate a graph  of people in the dataset, where an edge between two people, if they have ever visited the same cell tower, independantly of whether they were there at the same time. This produced the plot below:

This graph shows people connected based on visiting the same cell towers, but not necessarily at the same time. (full size)

The two plots show the same data, but with well connected node 22 removed on the right. This data implies that most people are using the same cell towers at some points in the dataset, which is promising for using cell tower a  location metric. However, this does not show any information about whether nodes were connected at the same time.

To further clarify the data, I processed the information provided for cell towers, which is stored in two columns in the dataset – oid and name, for example:

OID Name
35 40342, T-Mobile
36 40772, T-Mobile
38 0,
39 883, AT&T Wirel
40 2353, AT&T Wirel

I seperated out the cell ids and carrier names  into two columns, which enabled me to get the distribution of carrier names as below:

This shows the number of cells for each known carrier reported in the dataset (30 Others refers to 30 distinct reported carrier names, who’s sum totat was 429 reported towers) (full size)

This suggests that many of the towers are provided by different carriers, so this suggests that, we might have distinct clusters of users, that for some reason are connected by a few – perhaps they ‘roamed’ onto a new network, or changed provider at some point. This needs further clarification, and may only be able to be measured using ‘brute force’, by constructing the graph of people connected based on c0-locations of cell towers.

So far, the analysis is inconclusive, which means it will need a little more investigation. However, my feeling is that we WILL have enough overlap in the dataset to make it feasible to infer location from cell towers records.

Categories: Uncategorized
  1. No comments yet.
  1. No trackbacks yet.