Image tag relationship study

Abstract

Due to the increasing use of the Web, there is an increase in the amount of data. Tags which are used to

describe a subject it was tagged to, are also being widely used as a result. With a large amount of growing

data, it is essential to be able to search efficiently for useful information that the users need. Tag classes

become very important for this purpose by quickly filtering out unwanted information and reducing the

search space. It is also useful for other purposes such as recommendation of tags to the users. In this

project, tags studied were used on images and photos only, and were mainly used in photo sharing

websites such as flickr.com. The aim of this project was to implement a Java program to automatically

identify the tag class of a tag, and categorize tags into their respective classes. Both rulebased

classification methods which used boolean matching on a library of keywords, as well as modelbased

classification methods which used machine learning techniques were used to identify different sets of tag

classes. The finished program was run and tested against a set of 500 manually classified tags and the

results showed that it was able to classify most image tags correctly.