Acquiring structured knowledge from Flickr

Alex J Thomson


Supervised by Steven Schockaert; Moderated by Alun D Preece

Association rule mining is a popular technique to identify correlations in large amounts of data. A well-known example is market basket analysis, which identifies rules such as "people who buy bread and eggs usually also buy milk" from the buying behaviour of customers. This kind of knowledge can subsequently be used to decide how to organise a shop, or which products to reduce in price, among others.

Flickr is an online website for sharing photos, where people can add short textual descriptions, called tags, to the photos they are uploading. From the tagging behaviour of users, we may try to identify knowledge about different places, e.g. "if 'Spain' and 'beach' occur as tags, then usually also 'sun' occurs". In other words, we can use association rule mining to acquire knowledge from Flickr, by letting the tags that people assign to their photos take the role of items that are bought in market basket analysis. The resulting knowledge could then be used to improve the quality of location-based services. For instance, knowing in which parts of the world there are sunny beaches may help us to implement a travel recommendation engine.

The goal of this project is to implement a standard association rule mining

Initial Plan (19/10/2012) [Zip Archive]

Interim Report (14/12/2012) [Zip Archive]

Final Report (03/05/2013) [Zip Archive]

Publication Form