Methods to Study Flickr Users Behaviors

A couple of papers studying users behaviors in the context of online photo-sharing. Each has a specific angle on the data: social science (individual and group cooperation practices), data mining (who is looking at a photo) and visualization (understand the notion of place). I use them to inform the analysis of Flickr users practices to georeference and geotag their photos.

Prieur, C., Cardon, D., Beuscart, J.-S., Pissard, N., and Pons, P. (2008). The stength of weak cooperation: A case study on flickr. CoRR, abs/0802.2317.

Objective: Detail the concept of weak cooperation showing the great variety of uses (stockpiling, social media use, myspace like).
Data: 5M users, 150M photos, users, contacts, groups photos, comments, tags and favorites.
Methods: Distribution of flickr functionalities (contacts, comments, favorites); principal component analysis on correlation matrix to represent graphically the relations among nb photos, nb contacts (in/out), nb favorites (in/out); distribution of number of members and photos among Flickr groups, social graph with members of groups as nodes and proximity function between users from the similarity of the tags they use; define the social density of a group from the density of the social graph.

van Zwol, R. (2007). Flickr: Who is looking? In 2007 IEEE / WIC / ACM International Conference on Web Intelligence, pages 184–190. IEEE Computer Society.

Objective: Characterize user behaviors on temporal, social and spatial dimensions.
Data: 1.83M uploaded during 10 days and viewed over a period of 50 days.
Methods: Use a photo’s popularity to identify social networking and photo pooling. Distribution of photos views per bucket (a bucket groups the photos based on the number of views) 0-10%, 10-20%, 20-30%, 30-40%, 40-50% and > 50%. Same for the temporal dimension slices of time, and social dimension (comments, contacts, pools). Geographic distribution defined with two standard deviations (for the longitude and latitude) and a single value representing the geographic spreading of the photos views with the euclidian distance between the two standard deviations. This value is then applied to the slices.

Dykes, J., Purves, R., Edwardes, A. J., and Wood, J. (2008). Exploring volunteered geographic information to describe place: Visualization of the ’geograph british isles’ collection. In GIS Research UK (GISRUK 2008). Manchester Metropolitan University.

Objective: Understand the way language is used to describe landscape with geotagged photos
Data: 340,000 photographs with titles, comments and other metadata, georeference with at least 1km precision
Methods: Interactive and spatial treemaps of terms. Forcus on scene types (e.g. beach, village, mountain, hill) and scene type descriptors. Chi statistic maps of national trends for the selected combination of scene types. Spatial tag clouds to explore local variation in the terms.

Relation to my thesis:  Quantitative methods to understand user practices with online photo-sharing platforms in extensions to works such as Why We Tag: Motivations for Annotation in Mobile and Online Media