Exploring City Structure from Georeferenced Photos Using Graph - - PowerPoint PPT Presentation

Exploring City Structure from Georeferenced Photos Using Graph Centrality Measures

Katerina Vrotsou1, Natalia Andrienko1, Gennady Andrienko1, and Piotr Jankowski2

1Fraunhofer Institute IAIS 2San Diego State University

Appendix to demonstration paper:

submitted to: European Conference on Machine Learning and Principles and Practice

• f Knowledge Discovery in Databases (ECML PKDD) 2011

Sample of an

Exploratory analysis of Seattle Metropolitan Area using Flickr Photos

• We consider a Flickr dataset:

– From the Puget Sound metropolitan area, Washington State, USA (including the cities of Seattle, Bellevue, Kirkland, Redmond and neighbouring communities) – 577,053 photos – 9,324 photographers – January 1, 2005 - August 31, 2009

Graph of moves between places

• We retrieve a directed weighted graph

consisting of:

– 1,446 vertices (places) – 40,627 edges (moves) – Each edge is weighted by the number of moves between places

• and compute centrality scores for each of the

vertices (places)

Visual representation of moves between places

• Arrows represent

the movements between places

• They are rendered

with 40% transparency

• No structure can be

detected

Centrality Scores of Places

• In-degree
• Out-degree
• Weighted in-degree
• Weighted out-degree
• Closeness
• Betweenness
• Clustering coefficient

In- and out-degree

• In and out degree reveal an

almost identical picture.

• This can also be seen in

their scatter plot

In-degree

• ut-degree

In- and out-degrees reveal two core clusters of places in Seattle

2 1

1st core cluster of places

Connects the Westlake Center and Pike Place Market with the Pacific Science Center and Capitol Hill area around the Broadway This cluster of places is no surprise as it contains the main attractions of downtown Seattle which attract most of the city visitors.

Westlake Center and Pike Place Market Pacific Science Center Capitol Hill area

2nd core cluster of places

Located along the waterway spanning the urban villages of Freemont and Ballard With Gas Works Park having the highest in- and out-degree value in this cluster.

Ballard Freemont Gas Works Park

This cluster is less obvious and certainly the Gas Works Park area which has the highest value can be considered as being away from the city center. For this reason we analyze the corresponding time series of the area to see whether their high in and out degree stays over time.

In- and out-degrees of Gas Works Park

• ver time

In-degree time graph Out-degree time graph

Both graphs show high peaks during the summer months

Closeness centrality

Parallel coordinates plot

• f centrality

measures coloured by the closeness classes

High closeness values

High closeness values reveal an interesting picture of the core and periphery of well connected locations that were photographed.

1 2 4 5 3

• 1. The urban core including downtown Seattle along

with the Capitol Hill and Queen Anne Hill areas

• 2. The core is well connected to the areas to the north-

west (Ballard and Freemont), north (Green Lake) and north-east (University District) of downtown Seattle

The periphery is comprised of:

• 3. the areas on the east side of

Lake Washington (Bellevue, Kirkland, Redmond)

• 4. West Seattle with Alki Point

and Hamilton View Park

• 5. the Boeing Fields and Museum
• f Flight in South Seattle

4 5 3

Clustering w.r.t. centrality scores

Parallel coordinates plot of clusters including the most central places.

Clusters of higher scoring places

Clusters with high centrality scores

Clusters of high centrality areas include Kirkland (2 water front Parks) and Bellevue town center with the Bellevue shopping mall (upper left map), downtown area of Westlake Center, Freemont and Ballard along the waterway from Lake Union to Eliot Bay, Univ. of Washington campus, Capitol Hill, and the West Seattle’s Hamilton Viewpoint Part (upper right map), Gas Works Parks, Pacific Science Center and Pike Place Market area (lower maps)

Diagrams of parallel bars of centrality scores per area overlaid on the map

Places with low degrees are filtered out

Places with high in- and out-degrees and low weighted in- and out- degrees are filtered out… …resulting in areas that were visited frequently but mostly from the same few locations.

Filtering result

Bellevue shopping center

Moves to/from Bellevue shopping center

To From

The movement pattern to and from downtown Bellevue shopping center makes perfect sense as this shopping area is well known for its hip and chic stores.

This appendix presents a sample of a much larger exploratory analysis driven interactively by the analyst

The results of such analysis can be used in:

• City planning
• Transportation research
• Tourism planning and development
• Marketing and advertising